5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
87 /* At least one required character in the target string is expressible only in
89 static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
93 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
104 /* Valid for non-utf8 strings: avoids the reginclass
105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
108 : ANYOF_BITMAP_TEST(p,*(c)))
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
115 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
117 #define HOPc(pos,off) \
118 (char *)(reginfo->is_utf8_target \
119 ? reghop3((U8*)pos, off, \
120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
122 #define HOPBACKc(pos, off) \
123 (char*)(reginfo->is_utf8_target \
124 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
125 : (pos - off >= reginfo->strbeg) \
129 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
130 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
133 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
134 #define NEXTCHR_IS_EOS (nextchr < 0)
136 #define SET_nextchr \
137 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
139 #define SET_locinput(p) \
144 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \
146 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
147 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
148 1, 0, NULL, &flags); \
153 /* If in debug mode, we test that a known character properly matches */
155 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
157 utf8_char_in_property) \
158 LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \
159 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
161 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
163 utf8_char_in_property) \
164 LOAD_UTF8_CHARCLASS(swash_ptr, property_name)
167 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
168 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
169 swash_property_names[_CC_WORDCHAR], \
170 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
172 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
174 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
175 "_X_regular_begin", \
176 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
177 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
179 COMBINING_GRAVE_ACCENT_UTF8); \
182 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
183 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
185 /* for use after a quantifier and before an EXACT-like node -- japhy */
186 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
188 * NOTE that *nothing* that affects backtracking should be in here, specifically
189 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
190 * node that is in between two EXACT like nodes when ascertaining what the required
191 * "follow" character is. This should probably be moved to regex compile time
192 * although it may be done at run time beause of the REF possibility - more
193 * investigation required. -- demerphq
195 #define JUMPABLE(rn) ( \
197 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
199 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
200 OP(rn) == PLUS || OP(rn) == MINMOD || \
202 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
204 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
206 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
209 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
210 we don't need this definition. */
211 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
212 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
213 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
216 /* ... so we use this as its faster. */
217 #define IS_TEXT(rn) ( OP(rn)==EXACT )
218 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
219 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
220 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
225 Search for mandatory following text node; for lookahead, the text must
226 follow but for lookbehind (rn->flags != 0) we skip to the next step.
228 #define FIND_NEXT_IMPT(rn) STMT_START { \
229 while (JUMPABLE(rn)) { \
230 const OPCODE type = OP(rn); \
231 if (type == SUSPEND || PL_regkind[type] == CURLY) \
232 rn = NEXTOPER(NEXTOPER(rn)); \
233 else if (type == PLUS) \
235 else if (type == IFMATCH) \
236 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
237 else rn += NEXT_OFF(rn); \
241 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
242 * These are for the pre-composed Hangul syllables, which are all in a
243 * contiguous block and arranged there in such a way so as to facilitate
244 * alorithmic determination of their characteristics. As such, they don't need
245 * a swash, but can be determined by simple arithmetic. Almost all are
246 * GCB=LVT, but every 28th one is a GCB=LV */
247 #define SBASE 0xAC00 /* Start of block */
248 #define SCount 11172 /* Length of block */
251 #define SLAB_FIRST(s) (&(s)->states[0])
252 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
254 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
255 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
256 static regmatch_state * S_push_slab(pTHX);
258 #define REGCP_PAREN_ELEMS 3
259 #define REGCP_OTHER_ELEMS 3
260 #define REGCP_FRAME_ELEMS 1
261 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
262 * are needed for the regexp context stack bookkeeping. */
265 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
268 const int retval = PL_savestack_ix;
269 const int paren_elems_to_push =
270 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
271 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
272 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
274 GET_RE_DEBUG_FLAGS_DECL;
276 PERL_ARGS_ASSERT_REGCPPUSH;
278 if (paren_elems_to_push < 0)
279 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
280 paren_elems_to_push);
282 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
283 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
284 " out of range (%lu-%ld)",
286 (unsigned long)maxopenparen,
289 SSGROW(total_elems + REGCP_FRAME_ELEMS);
292 if ((int)maxopenparen > (int)parenfloor)
293 PerlIO_printf(Perl_debug_log,
294 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
299 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
300 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
301 SSPUSHIV(rex->offs[p].end);
302 SSPUSHIV(rex->offs[p].start);
303 SSPUSHINT(rex->offs[p].start_tmp);
304 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
305 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
307 (IV)rex->offs[p].start,
308 (IV)rex->offs[p].start_tmp,
312 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
313 SSPUSHINT(maxopenparen);
314 SSPUSHINT(rex->lastparen);
315 SSPUSHINT(rex->lastcloseparen);
316 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
321 /* These are needed since we do not localize EVAL nodes: */
322 #define REGCP_SET(cp) \
324 PerlIO_printf(Perl_debug_log, \
325 " Setting an EVAL scope, savestack=%"IVdf"\n", \
326 (IV)PL_savestack_ix)); \
329 #define REGCP_UNWIND(cp) \
331 if (cp != PL_savestack_ix) \
332 PerlIO_printf(Perl_debug_log, \
333 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
334 (IV)(cp), (IV)PL_savestack_ix)); \
337 #define UNWIND_PAREN(lp, lcp) \
338 for (n = rex->lastparen; n > lp; n--) \
339 rex->offs[n].end = -1; \
340 rex->lastparen = n; \
341 rex->lastcloseparen = lcp;
345 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
350 GET_RE_DEBUG_FLAGS_DECL;
352 PERL_ARGS_ASSERT_REGCPPOP;
354 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
356 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
357 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
358 rex->lastcloseparen = SSPOPINT;
359 rex->lastparen = SSPOPINT;
360 *maxopenparen_p = SSPOPINT;
362 i -= REGCP_OTHER_ELEMS;
363 /* Now restore the parentheses context. */
365 if (i || rex->lastparen + 1 <= rex->nparens)
366 PerlIO_printf(Perl_debug_log,
367 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
372 paren = *maxopenparen_p;
373 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
375 rex->offs[paren].start_tmp = SSPOPINT;
376 rex->offs[paren].start = SSPOPIV;
378 if (paren <= rex->lastparen)
379 rex->offs[paren].end = tmps;
380 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
381 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
383 (IV)rex->offs[paren].start,
384 (IV)rex->offs[paren].start_tmp,
385 (IV)rex->offs[paren].end,
386 (paren > rex->lastparen ? "(skipped)" : ""));
391 /* It would seem that the similar code in regtry()
392 * already takes care of this, and in fact it is in
393 * a better location to since this code can #if 0-ed out
394 * but the code in regtry() is needed or otherwise tests
395 * requiring null fields (pat.t#187 and split.t#{13,14}
396 * (as of patchlevel 7877) will fail. Then again,
397 * this code seems to be necessary or otherwise
398 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
399 * --jhi updated by dapm */
400 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
401 if (i > *maxopenparen_p)
402 rex->offs[i].start = -1;
403 rex->offs[i].end = -1;
404 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
405 " \\%"UVuf": %s ..-1 undeffing\n",
407 (i > *maxopenparen_p) ? "-1" : " "
413 /* restore the parens and associated vars at savestack position ix,
414 * but without popping the stack */
417 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
419 I32 tmpix = PL_savestack_ix;
420 PL_savestack_ix = ix;
421 regcppop(rex, maxopenparen_p);
422 PL_savestack_ix = tmpix;
425 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
428 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
430 /* Returns a boolean as to whether or not 'character' is a member of the
431 * Posix character class given by 'classnum' that should be equivalent to a
432 * value in the typedef '_char_class_number'.
434 * Ideally this could be replaced by a just an array of function pointers
435 * to the C library functions that implement the macros this calls.
436 * However, to compile, the precise function signatures are required, and
437 * these may vary from platform to to platform. To avoid having to figure
438 * out what those all are on each platform, I (khw) am using this method,
439 * which adds an extra layer of function call overhead (unless the C
440 * optimizer strips it away). But we don't particularly care about
441 * performance with locales anyway. */
443 switch ((_char_class_number) classnum) {
444 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
445 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
446 case _CC_ENUM_ASCII: return isASCII_LC(character);
447 case _CC_ENUM_BLANK: return isBLANK_LC(character);
448 case _CC_ENUM_CASED: return isLOWER_LC(character)
449 || isUPPER_LC(character);
450 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
451 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
452 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
453 case _CC_ENUM_LOWER: return isLOWER_LC(character);
454 case _CC_ENUM_PRINT: return isPRINT_LC(character);
455 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
456 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
457 case _CC_ENUM_SPACE: return isSPACE_LC(character);
458 case _CC_ENUM_UPPER: return isUPPER_LC(character);
459 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
460 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
461 default: /* VERTSPACE should never occur in locales */
462 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
465 assert(0); /* NOTREACHED */
470 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
472 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
473 * 'character' is a member of the Posix character class given by 'classnum'
474 * that should be equivalent to a value in the typedef
475 * '_char_class_number'.
477 * This just calls isFOO_lc on the code point for the character if it is in
478 * the range 0-255. Outside that range, all characters avoid Unicode
479 * rules, ignoring any locale. So use the Unicode function if this class
480 * requires a swash, and use the Unicode macro otherwise. */
482 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
484 if (UTF8_IS_INVARIANT(*character)) {
485 return isFOO_lc(classnum, *character);
487 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
488 return isFOO_lc(classnum,
489 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
492 if (classnum < _FIRST_NON_SWASH_CC) {
494 /* Initialize the swash unless done already */
495 if (! PL_utf8_swash_ptrs[classnum]) {
496 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
497 PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8",
498 swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags);
501 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
503 TRUE /* is UTF */ ));
506 switch ((_char_class_number) classnum) {
508 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
510 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
511 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
512 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
513 default: return 0; /* Things like CNTRL are always
517 assert(0); /* NOTREACHED */
522 * pregexec and friends
525 #ifndef PERL_IN_XSUB_RE
527 - pregexec - match a regexp against a string
530 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
531 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
532 /* stringarg: the point in the string at which to begin matching */
533 /* strend: pointer to null at end of string */
534 /* strbeg: real beginning of string */
535 /* minend: end of match must be >= minend bytes after stringarg. */
536 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
537 * itself is accessed via the pointers above */
538 /* nosave: For optimizations. */
540 PERL_ARGS_ASSERT_PREGEXEC;
543 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
544 nosave ? 0 : REXEC_COPY_STR);
549 * Need to implement the following flags for reg_anch:
551 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
553 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
554 * INTUIT_AUTORITATIVE_ML
555 * INTUIT_ONCE_NOML - Intuit can match in one location only.
558 * Another flag for this function: SECOND_TIME (so that float substrs
559 * with giant delta may be not rechecked).
562 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
563 Otherwise, only SvCUR(sv) is used to get strbeg. */
565 /* XXXX Some places assume that there is a fixed substring.
566 An update may be needed if optimizer marks as "INTUITable"
567 RExen without fixed substrings. Similarly, it is assumed that
568 lengths of all the strings are no more than minlen, thus they
569 cannot come from lookahead.
570 (Or minlen should take into account lookahead.)
571 NOTE: Some of this comment is not correct. minlen does now take account
572 of lookahead/behind. Further research is required. -- demerphq
576 /* A failure to find a constant substring means that there is no need to make
577 an expensive call to REx engine, thus we celebrate a failure. Similarly,
578 finding a substring too deep into the string means that fewer calls to
579 regtry() should be needed.
581 REx compiler's optimizer found 4 possible hints:
582 a) Anchored substring;
584 c) Whether we are anchored (beginning-of-line or \G);
585 d) First node (of those at offset 0) which may distinguish positions;
586 We use a)b)d) and multiline-part of c), and try to find a position in the
587 string which does not contradict any of them.
590 /* Most of decisions we do here should have been done at compile time.
591 The nodes of the REx which we used for the search should have been
592 deleted from the finite automaton. */
595 * rx: the regex to match against
596 * sv: the SV being matched: only used for utf8 flag; the string
597 * itself is accessed via the pointers below. Note that on
598 * something like an overloaded SV, SvPOK(sv) may be false
599 * and the string pointers may point to something unrelated to
601 * strbeg: real beginning of string
602 * strpos: the point in the string at which to begin matching
603 * strend: pointer to the byte following the last char of the string
604 * flags currently unused; set to 0
605 * data: currently unused; set to NULL
609 Perl_re_intuit_start(pTHX_
612 const char * const strbeg,
616 re_scream_pos_data *data)
619 struct regexp *const prog = ReANY(rx);
620 SSize_t start_shift = 0;
621 /* Should be nonnegative! */
622 SSize_t end_shift = 0;
626 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
628 char *other_last = NULL; /* other substr checked before this */
629 char *check_at = NULL; /* check substr found at this pos */
630 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
631 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
632 RXi_GET_DECL(prog,progi);
633 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
634 regmatch_info *const reginfo = ®info_buf;
636 const char * const i_strpos = strpos;
638 GET_RE_DEBUG_FLAGS_DECL;
640 PERL_ARGS_ASSERT_RE_INTUIT_START;
641 PERL_UNUSED_ARG(flags);
642 PERL_UNUSED_ARG(data);
644 /* CHR_DIST() would be more correct here but it makes things slow. */
645 if (prog->minlen > strend - strpos) {
646 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
647 "String too short... [re_intuit_start]\n"));
651 reginfo->is_utf8_target = cBOOL(utf8_target);
652 reginfo->info_aux = NULL;
653 reginfo->strbeg = strbeg;
654 reginfo->strend = strend;
655 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
657 /* not actually used within intuit, but zero for safety anyway */
658 reginfo->poscache_maxiter = 0;
661 if (!prog->check_utf8 && prog->check_substr)
662 to_utf8_substr(prog);
663 check = prog->check_utf8;
665 if (!prog->check_substr && prog->check_utf8) {
666 if (! to_byte_substr(prog)) {
667 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
670 check = prog->check_substr;
672 if (prog->extflags & RXf_ANCH) { /* Match at \G, beg-of-str or after \n */
673 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
674 || ( (prog->extflags & RXf_ANCH_BOL)
675 && !multiline ) ); /* Check after \n? */
678 /* we are only allowed to match at BOS or \G */
680 if (prog->extflags & RXf_ANCH_GPOS) {
681 /* in this case, we hope(!) that the caller has already
682 * set strpos to pos()-gofs, and will already have checked
683 * that this anchor position is legal
687 else if (!(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
688 && (strpos != strbeg))
690 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
693 if (prog->check_offset_min == prog->check_offset_max
694 && !(prog->extflags & RXf_CANY_SEEN)
695 && ! multiline) /* /m can cause \n's to match that aren't
696 accounted for in the string max length.
697 See [perl #115242] */
699 /* Substring at constant offset from beg-of-str... */
702 s = HOP3c(strpos, prog->check_offset_min, strend);
705 slen = SvCUR(check); /* >= 1 */
707 if ( strend - s > slen || strend - s < slen - 1
708 || (strend - s == slen && strend[-1] != '\n')) {
709 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
712 /* Now should match s[0..slen-2] */
714 if (slen && (*SvPVX_const(check) != *s
716 && memNE(SvPVX_const(check), s, slen)))) {
718 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
722 else if (*SvPVX_const(check) != *s
723 || ((slen = SvCUR(check)) > 1
724 && memNE(SvPVX_const(check), s, slen)))
727 goto success_at_start;
730 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
732 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
733 end_shift = prog->check_end_shift;
736 const SSize_t end = prog->check_offset_max + CHR_SVLEN(check)
737 - (SvTAIL(check) != 0);
738 const SSize_t eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
740 if (end_shift < eshift)
744 else { /* Can match at random position */
747 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
748 end_shift = prog->check_end_shift;
750 /* end shift should be non negative here */
753 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
755 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
756 (IV)end_shift, RX_PRECOMP(prog));
760 /* Find a possible match in the region s..strend by looking for
761 the "check" substring in the region corrected by start/end_shift. */
764 SSize_t srch_start_shift = start_shift;
765 SSize_t srch_end_shift = end_shift;
768 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
769 srch_end_shift -= ((strbeg - s) - srch_start_shift);
770 srch_start_shift = strbeg - s;
772 DEBUG_OPTIMISE_MORE_r({
773 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
774 (IV)prog->check_offset_min,
775 (IV)srch_start_shift,
777 (IV)prog->check_end_shift);
780 if (prog->extflags & RXf_CANY_SEEN) {
781 start_point= (U8*)(s + srch_start_shift);
782 end_point= (U8*)(strend - srch_end_shift);
784 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
785 end_point= HOP3(strend, -srch_end_shift, strbeg);
787 DEBUG_OPTIMISE_MORE_r({
788 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
789 (int)(end_point - start_point),
790 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
794 s = fbm_instr( start_point, end_point,
795 check, multiline ? FBMrf_MULTILINE : 0);
797 /* Update the count-of-usability, remove useless subpatterns,
801 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
802 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
803 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
804 (s ? "Found" : "Did not find"),
805 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
806 ? "anchored" : "floating"),
809 (s ? " at offset " : "...\n") );
814 /* Finish the diagnostic message */
815 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
817 /* XXX dmq: first branch is for positive lookbehind...
818 Our check string is offset from the beginning of the pattern.
819 So we need to do any stclass tests offset forward from that
828 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
829 Start with the other substr.
830 XXXX no SCREAM optimization yet - and a very coarse implementation
831 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
832 *always* match. Probably should be marked during compile...
833 Probably it is right to do no SCREAM here...
836 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
837 : (prog->float_substr && prog->anchored_substr))
839 /* Take into account the "other" substring. */
840 /* XXXX May be hopelessly wrong for UTF... */
843 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
846 char * const last = HOP3c(s, -start_shift, strbeg);
848 char * const saved_s = s;
851 t = s - prog->check_offset_max;
852 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
854 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
859 t = HOP3c(t, prog->anchored_offset, strend);
860 if (t < other_last) /* These positions already checked */
862 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
865 /* XXXX It is not documented what units *_offsets are in.
866 We assume bytes, but this is clearly wrong.
867 Meaning this code needs to be carefully reviewed for errors.
871 /* On end-of-str: see comment below. */
872 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
873 if (must == &PL_sv_undef) {
875 DEBUG_r(must = prog->anchored_utf8); /* for debug */
880 HOP3(HOP3(last1, prog->anchored_offset, strend)
881 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
883 multiline ? FBMrf_MULTILINE : 0
886 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
887 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
888 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
889 (s ? "Found" : "Contradicts"),
890 quoted, RE_SV_TAIL(must));
895 if (last1 >= last2) {
896 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
897 ", giving up...\n"));
900 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
901 ", trying floating at offset %ld...\n",
902 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
903 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
904 s = HOP3c(last, 1, strend);
908 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
909 (long)(s - i_strpos)));
910 t = HOP3c(s, -prog->anchored_offset, strbeg);
911 other_last = HOP3c(s, 1, strend);
919 else { /* Take into account the floating substring. */
921 char * const saved_s = s;
924 t = HOP3c(s, -start_shift, strbeg);
926 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
927 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
928 last = HOP3c(t, prog->float_max_offset, strend);
929 s = HOP3c(t, prog->float_min_offset, strend);
932 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
933 must = utf8_target ? prog->float_utf8 : prog->float_substr;
934 /* fbm_instr() takes into account exact value of end-of-str
935 if the check is SvTAIL(ed). Since false positives are OK,
936 and end-of-str is not later than strend we are OK. */
937 if (must == &PL_sv_undef) {
939 DEBUG_r(must = prog->float_utf8); /* for debug message */
942 s = fbm_instr((unsigned char*)s,
943 (unsigned char*)last + SvCUR(must)
945 must, multiline ? FBMrf_MULTILINE : 0);
947 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
948 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
949 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
950 (s ? "Found" : "Contradicts"),
951 quoted, RE_SV_TAIL(must));
955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
956 ", giving up...\n"));
959 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
960 ", trying anchored starting at offset %ld...\n",
961 (long)(saved_s + 1 - i_strpos)));
963 s = HOP3c(t, 1, strend);
967 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
968 (long)(s - i_strpos)));
969 other_last = s; /* Fix this later. --Hugo */
979 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
981 DEBUG_OPTIMISE_MORE_r(
982 PerlIO_printf(Perl_debug_log,
983 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
984 (IV)prog->check_offset_min,
985 (IV)prog->check_offset_max,
993 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
995 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
998 /* Fixed substring is found far enough so that the match
999 cannot start at strpos. */
1001 if (ml_anch && t[-1] != '\n') {
1002 /* Eventually fbm_*() should handle this, but often
1003 anchored_offset is not 0, so this check will not be wasted. */
1004 /* XXXX In the code below we prefer to look for "^" even in
1005 presence of anchored substrings. And we search even
1006 beyond the found float position. These pessimizations
1007 are historical artefacts only. */
1009 while (t < strend - prog->minlen) {
1011 if (t < check_at - prog->check_offset_min) {
1012 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1013 /* Since we moved from the found position,
1014 we definitely contradict the found anchored
1015 substr. Due to the above check we do not
1016 contradict "check" substr.
1017 Thus we can arrive here only if check substr
1018 is float. Redo checking for "other"=="fixed".
1021 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1022 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1023 goto do_other_anchored;
1025 /* We don't contradict the found floating substring. */
1026 /* XXXX Why not check for STCLASS? */
1028 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1029 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1032 /* Position contradicts check-string */
1033 /* XXXX probably better to look for check-string
1034 than for "\n", so one should lower the limit for t? */
1035 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1036 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1037 other_last = strpos = s = t + 1;
1042 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1043 PL_colors[0], PL_colors[1]));
1047 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1048 PL_colors[0], PL_colors[1]));
1052 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1055 /* The found string does not prohibit matching at strpos,
1056 - no optimization of calling REx engine can be performed,
1057 unless it was an MBOL and we are not after MBOL,
1058 or a future STCLASS check will fail this. */
1060 /* Even in this situation we may use MBOL flag if strpos is offset
1061 wrt the start of the string. */
1062 if (ml_anch && (strpos != strbeg) && strpos[-1] != '\n'
1063 /* May be due to an implicit anchor of m{.*foo} */
1064 && !(prog->intflags & PREGf_IMPLICIT))
1069 DEBUG_EXECUTE_r( if (ml_anch)
1070 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1071 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1074 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1076 prog->check_utf8 /* Could be deleted already */
1077 && --BmUSEFUL(prog->check_utf8) < 0
1078 && (prog->check_utf8 == prog->float_utf8)
1080 prog->check_substr /* Could be deleted already */
1081 && --BmUSEFUL(prog->check_substr) < 0
1082 && (prog->check_substr == prog->float_substr)
1085 /* If flags & SOMETHING - do not do it many times on the same match */
1086 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1087 /* XXX Does the destruction order has to change with utf8_target? */
1088 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1089 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1090 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1091 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1092 check = NULL; /* abort */
1094 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1095 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1096 if (prog->intflags & PREGf_IMPLICIT)
1097 prog->extflags &= ~RXf_ANCH_MBOL;
1098 /* XXXX This is a remnant of the old implementation. It
1099 looks wasteful, since now INTUIT can use many
1100 other heuristics. */
1101 prog->extflags &= ~RXf_USE_INTUIT;
1102 /* XXXX What other flags might need to be cleared in this branch? */
1108 /* Last resort... */
1109 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1110 /* trie stclasses are too expensive to use here, we are better off to
1111 leave it to regmatch itself */
1112 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1113 /* minlen == 0 is possible if regstclass is \b or \B,
1114 and the fixed substr is ''$.
1115 Since minlen is already taken into account, s+1 is before strend;
1116 accidentally, minlen >= 1 guaranties no false positives at s + 1
1117 even for \b or \B. But (minlen? 1 : 0) below assumes that
1118 regstclass does not come from lookahead... */
1119 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1120 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1121 const U8* const str = (U8*)STRING(progi->regstclass);
1122 /* XXX this value could be pre-computed */
1123 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1124 ? (reginfo->is_utf8_pat
1125 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1126 : STR_LEN(progi->regstclass))
1129 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1130 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1131 else if (prog->float_substr || prog->float_utf8)
1132 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1136 if (checked_upto < s)
1138 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1139 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1142 s = find_byclass(prog, progi->regstclass, checked_upto, endpos,
1148 const char *what = NULL;
1150 if (endpos == strend) {
1151 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1152 "Could not match STCLASS...\n") );
1155 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1156 "This position contradicts STCLASS...\n") );
1157 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1159 checked_upto = HOPBACKc(endpos, start_shift);
1160 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1161 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1162 /* Contradict one of substrings */
1163 if (prog->anchored_substr || prog->anchored_utf8) {
1164 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1165 DEBUG_EXECUTE_r( what = "anchored" );
1167 s = HOP3c(t, 1, strend);
1168 if (s + start_shift + end_shift > strend) {
1169 /* XXXX Should be taken into account earlier? */
1170 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1171 "Could not match STCLASS...\n") );
1176 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1177 "Looking for %s substr starting at offset %ld...\n",
1178 what, (long)(s + start_shift - i_strpos)) );
1181 /* Have both, check_string is floating */
1182 if (t + start_shift >= check_at) /* Contradicts floating=check */
1183 goto retry_floating_check;
1184 /* Recheck anchored substring, but not floating... */
1188 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1189 "Looking for anchored substr starting at offset %ld...\n",
1190 (long)(other_last - i_strpos)) );
1191 goto do_other_anchored;
1193 /* Another way we could have checked stclass at the
1194 current position only: */
1199 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1200 "Looking for /%s^%s/m starting at offset %ld...\n",
1201 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1204 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1206 /* Check is floating substring. */
1207 retry_floating_check:
1208 t = check_at - start_shift;
1209 DEBUG_EXECUTE_r( what = "floating" );
1210 goto hop_and_restart;
1213 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1214 "By STCLASS: moving %ld --> %ld\n",
1215 (long)(t - i_strpos), (long)(s - i_strpos))
1219 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1220 "Does not contradict STCLASS...\n");
1225 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1226 PL_colors[4], (check ? "Guessed" : "Giving up"),
1227 PL_colors[5], (long)(s - i_strpos)) );
1230 fail_finish: /* Substring not found */
1231 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1232 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1234 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1235 PL_colors[4], PL_colors[5]));
1239 #define DECL_TRIE_TYPE(scan) \
1240 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1241 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1242 trie_type = ((scan->flags == EXACT) \
1243 ? (utf8_target ? trie_utf8 : trie_plain) \
1244 : (scan->flags == EXACTFA) \
1245 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1246 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1248 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1251 U8 flags = FOLD_FLAGS_FULL; \
1252 switch (trie_type) { \
1253 case trie_utf8_exactfa_fold: \
1254 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1255 /* FALL THROUGH */ \
1256 case trie_utf8_fold: \
1257 if ( foldlen>0 ) { \
1258 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1263 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags, NULL); \
1264 len = UTF8SKIP(uc); \
1265 skiplen = UNISKIP( uvc ); \
1266 foldlen -= skiplen; \
1267 uscan = foldbuf + skiplen; \
1270 case trie_latin_utf8_exactfa_fold: \
1271 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1272 /* FALL THROUGH */ \
1273 case trie_latin_utf8_fold: \
1274 if ( foldlen>0 ) { \
1275 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1281 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1282 skiplen = UNISKIP( uvc ); \
1283 foldlen -= skiplen; \
1284 uscan = foldbuf + skiplen; \
1288 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1295 charid = trie->charmap[ uvc ]; \
1299 if (widecharmap) { \
1300 SV** const svpp = hv_fetch(widecharmap, \
1301 (char*)&uvc, sizeof(UV), 0); \
1303 charid = (U16)SvIV(*svpp); \
1308 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1312 && (ln == 1 || folder(s, pat_string, ln)) \
1313 && (reginfo->intuit || regtry(reginfo, &s)) )\
1319 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1321 while (s < strend) { \
1327 #define REXEC_FBC_SCAN(CoDe) \
1329 while (s < strend) { \
1335 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1336 REXEC_FBC_UTF8_SCAN( \
1338 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1347 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1350 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1359 #define REXEC_FBC_TRYIT \
1360 if ((reginfo->intuit || regtry(reginfo, &s))) \
1363 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1364 if (utf8_target) { \
1365 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1368 REXEC_FBC_CLASS_SCAN(CoNd); \
1371 #define DUMP_EXEC_POS(li,s,doutf8) \
1372 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1376 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1377 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1378 tmp = TEST_NON_UTF8(tmp); \
1379 REXEC_FBC_UTF8_SCAN( \
1380 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1389 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1390 if (s == reginfo->strbeg) { \
1394 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1395 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1398 LOAD_UTF8_CHARCLASS_ALNUM(); \
1399 REXEC_FBC_UTF8_SCAN( \
1400 if (tmp == ! (TeSt2_UtF8)) { \
1409 /* The only difference between the BOUND and NBOUND cases is that
1410 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1411 * NBOUND. This is accomplished by passing it in either the if or else clause,
1412 * with the other one being empty */
1413 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1414 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1416 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1417 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1419 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1420 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1422 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1423 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1426 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1427 * be passed in completely with the variable name being tested, which isn't
1428 * such a clean interface, but this is easier to read than it was before. We
1429 * are looking for the boundary (or non-boundary between a word and non-word
1430 * character. The utf8 and non-utf8 cases have the same logic, but the details
1431 * must be different. Find the "wordness" of the character just prior to this
1432 * one, and compare it with the wordness of this one. If they differ, we have
1433 * a boundary. At the beginning of the string, pretend that the previous
1434 * character was a new-line */
1435 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1436 if (utf8_target) { \
1439 else { /* Not utf8 */ \
1440 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1441 tmp = TEST_NON_UTF8(tmp); \
1443 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1452 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
1455 /* We know what class REx starts with. Try to find this position... */
1456 /* if reginfo->intuit, its a dryrun */
1457 /* annoyingly all the vars in this routine have different names from their counterparts
1458 in regmatch. /grrr */
1461 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1462 const char *strend, regmatch_info *reginfo)
1465 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1466 char *pat_string; /* The pattern's exactish string */
1467 char *pat_end; /* ptr to end char of pat_string */
1468 re_fold_t folder; /* Function for computing non-utf8 folds */
1469 const U8 *fold_array; /* array for folding ords < 256 */
1475 I32 tmp = 1; /* Scratch variable? */
1476 const bool utf8_target = reginfo->is_utf8_target;
1477 UV utf8_fold_flags = 0;
1478 const bool is_utf8_pat = reginfo->is_utf8_pat;
1479 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1480 with a result inverts that result, as 0^1 =
1482 _char_class_number classnum;
1484 RXi_GET_DECL(prog,progi);
1486 PERL_ARGS_ASSERT_FIND_BYCLASS;
1488 /* We know what class it must start with. */
1491 case ANYOF_SYNTHETIC:
1492 case ANYOF_WARN_SUPER:
1494 REXEC_FBC_UTF8_CLASS_SCAN(
1495 reginclass(prog, c, (U8*)s, utf8_target));
1498 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1503 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1510 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1511 assert(! is_utf8_pat);
1514 if (is_utf8_pat || utf8_target) {
1515 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1516 goto do_exactf_utf8;
1518 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1519 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1520 goto do_exactf_non_utf8; /* isn't dealt with by these */
1522 case EXACTF: /* This node only generated for non-utf8 patterns */
1523 assert(! is_utf8_pat);
1525 utf8_fold_flags = 0;
1526 goto do_exactf_utf8;
1528 fold_array = PL_fold;
1530 goto do_exactf_non_utf8;
1533 if (is_utf8_pat || utf8_target) {
1534 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1535 goto do_exactf_utf8;
1537 fold_array = PL_fold_locale;
1538 folder = foldEQ_locale;
1539 goto do_exactf_non_utf8;
1543 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1545 goto do_exactf_utf8;
1548 if (is_utf8_pat || utf8_target) {
1549 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1550 goto do_exactf_utf8;
1553 /* Any 'ss' in the pattern should have been replaced by regcomp,
1554 * so we don't have to worry here about this single special case
1555 * in the Latin1 range */
1556 fold_array = PL_fold_latin1;
1557 folder = foldEQ_latin1;
1561 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1562 are no glitches with fold-length differences
1563 between the target string and pattern */
1565 /* The idea in the non-utf8 EXACTF* cases is to first find the
1566 * first character of the EXACTF* node and then, if necessary,
1567 * case-insensitively compare the full text of the node. c1 is the
1568 * first character. c2 is its fold. This logic will not work for
1569 * Unicode semantics and the german sharp ss, which hence should
1570 * not be compiled into a node that gets here. */
1571 pat_string = STRING(c);
1572 ln = STR_LEN(c); /* length to match in octets/bytes */
1574 /* We know that we have to match at least 'ln' bytes (which is the
1575 * same as characters, since not utf8). If we have to match 3
1576 * characters, and there are only 2 availabe, we know without
1577 * trying that it will fail; so don't start a match past the
1578 * required minimum number from the far end */
1579 e = HOP3c(strend, -((SSize_t)ln), s);
1581 if (reginfo->intuit && e < s) {
1582 e = s; /* Due to minlen logic of intuit() */
1586 c2 = fold_array[c1];
1587 if (c1 == c2) { /* If char and fold are the same */
1588 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1591 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1599 /* If one of the operands is in utf8, we can't use the simpler folding
1600 * above, due to the fact that many different characters can have the
1601 * same fold, or portion of a fold, or different- length fold */
1602 pat_string = STRING(c);
1603 ln = STR_LEN(c); /* length to match in octets/bytes */
1604 pat_end = pat_string + ln;
1605 lnc = is_utf8_pat /* length to match in characters */
1606 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1609 /* We have 'lnc' characters to match in the pattern, but because of
1610 * multi-character folding, each character in the target can match
1611 * up to 3 characters (Unicode guarantees it will never exceed
1612 * this) if it is utf8-encoded; and up to 2 if not (based on the
1613 * fact that the Latin 1 folds are already determined, and the
1614 * only multi-char fold in that range is the sharp-s folding to
1615 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1616 * string character. Adjust lnc accordingly, rounding up, so that
1617 * if we need to match at least 4+1/3 chars, that really is 5. */
1618 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1619 lnc = (lnc + expansion - 1) / expansion;
1621 /* As in the non-UTF8 case, if we have to match 3 characters, and
1622 * only 2 are left, it's guaranteed to fail, so don't start a
1623 * match that would require us to go beyond the end of the string
1625 e = HOP3c(strend, -((SSize_t)lnc), s);
1627 if (reginfo->intuit && e < s) {
1628 e = s; /* Due to minlen logic of intuit() */
1631 /* XXX Note that we could recalculate e to stop the loop earlier,
1632 * as the worst case expansion above will rarely be met, and as we
1633 * go along we would usually find that e moves further to the left.
1634 * This would happen only after we reached the point in the loop
1635 * where if there were no expansion we should fail. Unclear if
1636 * worth the expense */
1639 char *my_strend= (char *)strend;
1640 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1641 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1642 && (reginfo->intuit || regtry(reginfo, &s)) )
1646 s += (utf8_target) ? UTF8SKIP(s) : 1;
1651 RXp_MATCH_TAINTED_on(prog);
1652 FBC_BOUND(isWORDCHAR_LC,
1653 isWORDCHAR_LC_uvchr(tmp),
1654 isWORDCHAR_LC_utf8((U8*)s));
1657 RXp_MATCH_TAINTED_on(prog);
1658 FBC_NBOUND(isWORDCHAR_LC,
1659 isWORDCHAR_LC_uvchr(tmp),
1660 isWORDCHAR_LC_utf8((U8*)s));
1663 FBC_BOUND(isWORDCHAR,
1664 isWORDCHAR_uni(tmp),
1665 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1668 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1670 isWORDCHAR_A((U8*)s));
1673 FBC_NBOUND(isWORDCHAR,
1674 isWORDCHAR_uni(tmp),
1675 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1678 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1680 isWORDCHAR_A((U8*)s));
1683 FBC_BOUND(isWORDCHAR_L1,
1684 isWORDCHAR_uni(tmp),
1685 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1688 FBC_NBOUND(isWORDCHAR_L1,
1689 isWORDCHAR_uni(tmp),
1690 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1693 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1694 is_LNBREAK_latin1_safe(s, strend)
1698 /* The argument to all the POSIX node types is the class number to pass to
1699 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1706 RXp_MATCH_TAINTED_on(prog);
1707 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1708 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1723 /* The complement of something that matches only ASCII matches all
1724 * UTF-8 variant code points, plus everything in ASCII that isn't
1726 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s)
1727 || ! _generic_isCC_A(*s, FLAGS(c)));
1736 /* Don't need to worry about utf8, as it can match only a single
1737 * byte invariant character. */
1738 REXEC_FBC_CLASS_SCAN(
1739 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1747 if (! utf8_target) {
1748 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1754 classnum = (_char_class_number) FLAGS(c);
1755 if (classnum < _FIRST_NON_SWASH_CC) {
1756 while (s < strend) {
1758 /* We avoid loading in the swash as long as possible, but
1759 * should we have to, we jump to a separate loop. This
1760 * extra 'if' statement is what keeps this code from being
1761 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1762 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1763 goto found_above_latin1;
1765 if ((UTF8_IS_INVARIANT(*s)
1766 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1768 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1769 && to_complement ^ cBOOL(
1770 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
1774 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1786 else switch (classnum) { /* These classes are implemented as
1788 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1789 revert the change of \v matching this */
1792 case _CC_ENUM_PSXSPC:
1793 REXEC_FBC_UTF8_CLASS_SCAN(
1794 to_complement ^ cBOOL(isSPACE_utf8(s)));
1797 case _CC_ENUM_BLANK:
1798 REXEC_FBC_UTF8_CLASS_SCAN(
1799 to_complement ^ cBOOL(isBLANK_utf8(s)));
1802 case _CC_ENUM_XDIGIT:
1803 REXEC_FBC_UTF8_CLASS_SCAN(
1804 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1807 case _CC_ENUM_VERTSPACE:
1808 REXEC_FBC_UTF8_CLASS_SCAN(
1809 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1812 case _CC_ENUM_CNTRL:
1813 REXEC_FBC_UTF8_CLASS_SCAN(
1814 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1818 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
1819 assert(0); /* NOTREACHED */
1824 found_above_latin1: /* Here we have to load a swash to get the result
1825 for the current code point */
1826 if (! PL_utf8_swash_ptrs[classnum]) {
1827 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1828 PL_utf8_swash_ptrs[classnum] =
1829 _core_swash_init("utf8", swash_property_names[classnum],
1830 &PL_sv_undef, 1, 0, NULL, &flags);
1833 /* This is a copy of the loop above for swash classes, though using the
1834 * FBC macro instead of being expanded out. Since we've loaded the
1835 * swash, we don't have to check for that each time through the loop */
1836 REXEC_FBC_UTF8_CLASS_SCAN(
1837 to_complement ^ cBOOL(_generic_utf8(
1840 swash_fetch(PL_utf8_swash_ptrs[classnum],
1848 /* what trie are we using right now */
1849 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1850 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1851 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1853 const char *last_start = strend - trie->minlen;
1855 const char *real_start = s;
1857 STRLEN maxlen = trie->maxlen;
1859 U8 **points; /* map of where we were in the input string
1860 when reading a given char. For ASCII this
1861 is unnecessary overhead as the relationship
1862 is always 1:1, but for Unicode, especially
1863 case folded Unicode this is not true. */
1864 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1868 GET_RE_DEBUG_FLAGS_DECL;
1870 /* We can't just allocate points here. We need to wrap it in
1871 * an SV so it gets freed properly if there is a croak while
1872 * running the match */
1875 sv_points=newSV(maxlen * sizeof(U8 *));
1876 SvCUR_set(sv_points,
1877 maxlen * sizeof(U8 *));
1878 SvPOK_on(sv_points);
1879 sv_2mortal(sv_points);
1880 points=(U8**)SvPV_nolen(sv_points );
1881 if ( trie_type != trie_utf8_fold
1882 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1885 bitmap=(U8*)trie->bitmap;
1887 bitmap=(U8*)ANYOF_BITMAP(c);
1889 /* this is the Aho-Corasick algorithm modified a touch
1890 to include special handling for long "unknown char" sequences.
1891 The basic idea being that we use AC as long as we are dealing
1892 with a possible matching char, when we encounter an unknown char
1893 (and we have not encountered an accepting state) we scan forward
1894 until we find a legal starting char.
1895 AC matching is basically that of trie matching, except that when
1896 we encounter a failing transition, we fall back to the current
1897 states "fail state", and try the current char again, a process
1898 we repeat until we reach the root state, state 1, or a legal
1899 transition. If we fail on the root state then we can either
1900 terminate if we have reached an accepting state previously, or
1901 restart the entire process from the beginning if we have not.
1904 while (s <= last_start) {
1905 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1913 U8 *uscan = (U8*)NULL;
1914 U8 *leftmost = NULL;
1916 U32 accepted_word= 0;
1920 while ( state && uc <= (U8*)strend ) {
1922 U32 word = aho->states[ state ].wordnum;
1926 DEBUG_TRIE_EXECUTE_r(
1927 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1928 dump_exec_pos( (char *)uc, c, strend, real_start,
1929 (char *)uc, utf8_target );
1930 PerlIO_printf( Perl_debug_log,
1931 " Scanning for legal start char...\n");
1935 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1939 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1945 if (uc >(U8*)last_start) break;
1949 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1950 if (!leftmost || lpos < leftmost) {
1951 DEBUG_r(accepted_word=word);
1957 points[pointpos++ % maxlen]= uc;
1958 if (foldlen || uc < (U8*)strend) {
1959 REXEC_TRIE_READ_CHAR(trie_type, trie,
1961 uscan, len, uvc, charid, foldlen,
1963 DEBUG_TRIE_EXECUTE_r({
1964 dump_exec_pos( (char *)uc, c, strend,
1965 real_start, s, utf8_target);
1966 PerlIO_printf(Perl_debug_log,
1967 " Charid:%3u CP:%4"UVxf" ",
1979 word = aho->states[ state ].wordnum;
1981 base = aho->states[ state ].trans.base;
1983 DEBUG_TRIE_EXECUTE_r({
1985 dump_exec_pos( (char *)uc, c, strend, real_start,
1987 PerlIO_printf( Perl_debug_log,
1988 "%sState: %4"UVxf", word=%"UVxf,
1989 failed ? " Fail transition to " : "",
1990 (UV)state, (UV)word);
1996 ( ((offset = base + charid
1997 - 1 - trie->uniquecharcount)) >= 0)
1998 && ((U32)offset < trie->lasttrans)
1999 && trie->trans[offset].check == state
2000 && (tmp=trie->trans[offset].next))
2002 DEBUG_TRIE_EXECUTE_r(
2003 PerlIO_printf( Perl_debug_log," - legal\n"));
2008 DEBUG_TRIE_EXECUTE_r(
2009 PerlIO_printf( Perl_debug_log," - fail\n"));
2011 state = aho->fail[state];
2015 /* we must be accepting here */
2016 DEBUG_TRIE_EXECUTE_r(
2017 PerlIO_printf( Perl_debug_log," - accepting\n"));
2026 if (!state) state = 1;
2029 if ( aho->states[ state ].wordnum ) {
2030 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2031 if (!leftmost || lpos < leftmost) {
2032 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2037 s = (char*)leftmost;
2038 DEBUG_TRIE_EXECUTE_r({
2040 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2041 (UV)accepted_word, (IV)(s - real_start)
2044 if (reginfo->intuit || regtry(reginfo, &s)) {
2050 DEBUG_TRIE_EXECUTE_r({
2051 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2054 DEBUG_TRIE_EXECUTE_r(
2055 PerlIO_printf( Perl_debug_log,"No match.\n"));
2064 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2072 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2073 * flags have same meanings as with regexec_flags() */
2076 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2083 struct regexp *const prog = ReANY(rx);
2085 if (flags & REXEC_COPY_STR) {
2089 PerlIO_printf(Perl_debug_log,
2090 "Copy on write: regexp capture, type %d\n",
2093 /* Create a new COW SV to share the match string and store
2094 * in saved_copy, unless the current COW SV in saved_copy
2095 * is valid and suitable for our purpose */
2096 if (( prog->saved_copy
2097 && SvIsCOW(prog->saved_copy)
2098 && SvPOKp(prog->saved_copy)
2101 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2103 /* just reuse saved_copy SV */
2104 if (RXp_MATCH_COPIED(prog)) {
2105 Safefree(prog->subbeg);
2106 RXp_MATCH_COPIED_off(prog);
2110 /* create new COW SV to share string */
2111 RX_MATCH_COPY_FREE(rx);
2112 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2114 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2115 assert (SvPOKp(prog->saved_copy));
2116 prog->sublen = strend - strbeg;
2117 prog->suboffset = 0;
2118 prog->subcoffset = 0;
2123 SSize_t max = strend - strbeg;
2126 if ( (flags & REXEC_COPY_SKIP_POST)
2127 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2128 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2129 ) { /* don't copy $' part of string */
2132 /* calculate the right-most part of the string covered
2133 * by a capture. Due to look-ahead, this may be to
2134 * the right of $&, so we have to scan all captures */
2135 while (n <= prog->lastparen) {
2136 if (prog->offs[n].end > max)
2137 max = prog->offs[n].end;
2141 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2142 ? prog->offs[0].start
2144 assert(max >= 0 && max <= strend - strbeg);
2147 if ( (flags & REXEC_COPY_SKIP_PRE)
2148 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2149 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2150 ) { /* don't copy $` part of string */
2153 /* calculate the left-most part of the string covered
2154 * by a capture. Due to look-behind, this may be to
2155 * the left of $&, so we have to scan all captures */
2156 while (min && n <= prog->lastparen) {
2157 if ( prog->offs[n].start != -1
2158 && prog->offs[n].start < min)
2160 min = prog->offs[n].start;
2164 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2165 && min > prog->offs[0].end
2167 min = prog->offs[0].end;
2171 assert(min >= 0 && min <= max && min <= strend - strbeg);
2174 if (RX_MATCH_COPIED(rx)) {
2175 if (sublen > prog->sublen)
2177 (char*)saferealloc(prog->subbeg, sublen+1);
2180 prog->subbeg = (char*)safemalloc(sublen+1);
2181 Copy(strbeg + min, prog->subbeg, sublen, char);
2182 prog->subbeg[sublen] = '\0';
2183 prog->suboffset = min;
2184 prog->sublen = sublen;
2185 RX_MATCH_COPIED_on(rx);
2187 prog->subcoffset = prog->suboffset;
2188 if (prog->suboffset && utf8_target) {
2189 /* Convert byte offset to chars.
2190 * XXX ideally should only compute this if @-/@+
2191 * has been seen, a la PL_sawampersand ??? */
2193 /* If there's a direct correspondence between the
2194 * string which we're matching and the original SV,
2195 * then we can use the utf8 len cache associated with
2196 * the SV. In particular, it means that under //g,
2197 * sv_pos_b2u() will use the previously cached
2198 * position to speed up working out the new length of
2199 * subcoffset, rather than counting from the start of
2200 * the string each time. This stops
2201 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2202 * from going quadratic */
2203 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2204 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2205 SV_GMAGIC|SV_CONST_RETURN);
2207 prog->subcoffset = utf8_length((U8*)strbeg,
2208 (U8*)(strbeg+prog->suboffset));
2212 RX_MATCH_COPY_FREE(rx);
2213 prog->subbeg = strbeg;
2214 prog->suboffset = 0;
2215 prog->subcoffset = 0;
2216 prog->sublen = strend - strbeg;
2224 - regexec_flags - match a regexp against a string
2227 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2228 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2229 /* stringarg: the point in the string at which to begin matching */
2230 /* strend: pointer to null at end of string */
2231 /* strbeg: real beginning of string */
2232 /* minend: end of match must be >= minend bytes after stringarg. */
2233 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2234 * itself is accessed via the pointers above */
2235 /* data: May be used for some additional optimizations.
2236 Currently unused. */
2237 /* flags: For optimizations. See REXEC_* in regexp.h */
2241 struct regexp *const prog = ReANY(rx);
2245 SSize_t minlen; /* must match at least this many chars */
2246 SSize_t dontbother = 0; /* how many characters not to try at end */
2247 const bool utf8_target = cBOOL(DO_UTF8(sv));
2249 RXi_GET_DECL(prog,progi);
2250 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2251 regmatch_info *const reginfo = ®info_buf;
2252 regexp_paren_pair *swap = NULL;
2254 GET_RE_DEBUG_FLAGS_DECL;
2256 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2257 PERL_UNUSED_ARG(data);
2259 /* Be paranoid... */
2260 if (prog == NULL || stringarg == NULL) {
2261 Perl_croak(aTHX_ "NULL regexp parameter");
2266 debug_start_match(rx, utf8_target, stringarg, strend,
2270 startpos = stringarg;
2272 if (prog->extflags & RXf_GPOS_SEEN) {
2275 /* set reginfo->ganch, the position where \G can match */
2278 (flags & REXEC_IGNOREPOS)
2279 ? stringarg /* use start pos rather than pos() */
2280 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2281 /* Defined pos(): */
2282 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2283 : strbeg; /* pos() not defined; use start of string */
2285 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2286 "GPOS ganch set to strbeg[%"IVdf"]\n", reginfo->ganch - strbeg));
2288 /* in the presence of \G, we may need to start looking earlier in
2289 * the string than the suggested start point of stringarg:
2290 * if prog->gofs is set, then that's a known, fixed minimum
2293 * /ab|c\G/: gofs = 1
2294 * or if the minimum offset isn't known, then we have to go back
2295 * to the start of the string, e.g. /w+\G/
2298 if (prog->extflags & RXf_ANCH_GPOS) {
2299 startpos = reginfo->ganch - prog->gofs;
2301 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2303 DEBUG_r(PerlIO_printf(Perl_debug_log,
2304 "fail: ganch-gofs before earliest possible start\n"));
2308 else if (prog->gofs) {
2309 if (startpos - prog->gofs < strbeg)
2312 startpos -= prog->gofs;
2314 else if (prog->extflags & RXf_GPOS_FLOAT)
2318 minlen = prog->minlen;
2319 if ((startpos + minlen) > strend || startpos < strbeg) {
2320 DEBUG_r(PerlIO_printf(Perl_debug_log,
2321 "Regex match can't succeed, so not even tried\n"));
2325 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2326 * which will call destuctors to reset PL_regmatch_state, free higher
2327 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2328 * regmatch_info_aux_eval */
2330 oldsave = PL_savestack_ix;
2334 if ((prog->extflags & RXf_USE_INTUIT)
2335 && !(flags & REXEC_CHECKED))
2337 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2342 if (prog->extflags & RXf_CHECK_ALL) {
2343 /* we can match based purely on the result of INTUIT.
2344 * Set up captures etc just for $& and $-[0]
2345 * (an intuit-only match wont have $1,$2,..) */
2346 assert(!prog->nparens);
2348 /* s/// doesn't like it if $& is earlier than where we asked it to
2349 * start searching (which can happen on something like /.\G/) */
2350 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2353 /* this should only be possible under \G */
2354 assert(prog->extflags & RXf_GPOS_SEEN);
2355 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2356 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2360 /* match via INTUIT shouldn't have any captures.
2361 * Let @-, @+, $^N know */
2362 prog->lastparen = prog->lastcloseparen = 0;
2363 RX_MATCH_UTF8_set(rx, utf8_target);
2364 prog->offs[0].start = s - strbeg;
2365 prog->offs[0].end = utf8_target
2366 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2367 : s - strbeg + prog->minlenret;
2368 if ( !(flags & REXEC_NOT_FIRST) )
2369 S_reg_set_capture_string(aTHX_ rx,
2371 sv, flags, utf8_target);
2377 multiline = prog->extflags & RXf_PMf_MULTILINE;
2379 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2380 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2381 "String too short [regexec_flags]...\n"));
2385 /* Check validity of program. */
2386 if (UCHARAT(progi->program) != REG_MAGIC) {
2387 Perl_croak(aTHX_ "corrupted regexp program");
2390 RX_MATCH_TAINTED_off(rx);
2392 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2393 reginfo->intuit = 0;
2394 reginfo->is_utf8_target = cBOOL(utf8_target);
2395 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2396 reginfo->warned = FALSE;
2397 reginfo->strbeg = strbeg;
2399 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2400 reginfo->strend = strend;
2401 /* see how far we have to get to not match where we matched before */
2402 reginfo->till = stringarg + minend;
2404 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv) && !IS_PADGV(sv)) {
2405 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2406 S_cleanup_regmatch_info_aux has executed (registered by
2407 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2408 magic belonging to this SV.
2409 Not newSVsv, either, as it does not COW.
2411 reginfo->sv = newSV(0);
2412 sv_setsv(reginfo->sv, sv);
2413 SAVEFREESV(reginfo->sv);
2416 /* reserve next 2 or 3 slots in PL_regmatch_state:
2417 * slot N+0: may currently be in use: skip it
2418 * slot N+1: use for regmatch_info_aux struct
2419 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2420 * slot N+3: ready for use by regmatch()
2424 regmatch_state *old_regmatch_state;
2425 regmatch_slab *old_regmatch_slab;
2426 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2428 /* on first ever match, allocate first slab */
2429 if (!PL_regmatch_slab) {
2430 Newx(PL_regmatch_slab, 1, regmatch_slab);
2431 PL_regmatch_slab->prev = NULL;
2432 PL_regmatch_slab->next = NULL;
2433 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2436 old_regmatch_state = PL_regmatch_state;
2437 old_regmatch_slab = PL_regmatch_slab;
2439 for (i=0; i <= max; i++) {
2441 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2443 reginfo->info_aux_eval =
2444 reginfo->info_aux->info_aux_eval =
2445 &(PL_regmatch_state->u.info_aux_eval);
2447 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2448 PL_regmatch_state = S_push_slab(aTHX);
2451 /* note initial PL_regmatch_state position; at end of match we'll
2452 * pop back to there and free any higher slabs */
2454 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2455 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2456 reginfo->info_aux->poscache = NULL;
2458 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2460 if ((prog->extflags & RXf_EVAL_SEEN))
2461 S_setup_eval_state(aTHX_ reginfo);
2463 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2466 /* If there is a "must appear" string, look for it. */
2468 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2469 /* We have to be careful. If the previous successful match
2470 was from this regex we don't want a subsequent partially
2471 successful match to clobber the old results.
2472 So when we detect this possibility we add a swap buffer
2473 to the re, and switch the buffer each match. If we fail,
2474 we switch it back; otherwise we leave it swapped.
2477 /* do we need a save destructor here for eval dies? */
2478 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2479 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2480 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2487 /* Simplest case: anchored match need be tried only once. */
2488 /* [unless only anchor is BOL and multiline is set] */
2489 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2490 if (s == startpos && regtry(reginfo, &s))
2492 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2493 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2498 dontbother = minlen - 1;
2499 end = HOP3c(strend, -dontbother, strbeg) - 1;
2500 /* for multiline we only have to try after newlines */
2501 if (prog->check_substr || prog->check_utf8) {
2502 /* because of the goto we can not easily reuse the macros for bifurcating the
2503 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2506 goto after_try_utf8;
2508 if (regtry(reginfo, &s)) {
2515 if (prog->extflags & RXf_USE_INTUIT) {
2516 s = re_intuit_start(rx, sv, strbeg,
2517 s + UTF8SKIP(s), strend, flags, NULL);
2526 } /* end search for check string in unicode */
2528 if (s == startpos) {
2529 goto after_try_latin;
2532 if (regtry(reginfo, &s)) {
2539 if (prog->extflags & RXf_USE_INTUIT) {
2540 s = re_intuit_start(rx, sv, strbeg,
2541 s + 1, strend, flags, NULL);
2550 } /* end search for check string in latin*/
2551 } /* end search for check string */
2552 else { /* search for newline */
2554 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2557 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2558 while (s <= end) { /* note it could be possible to match at the end of the string */
2559 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2560 if (regtry(reginfo, &s))
2564 } /* end search for newline */
2565 } /* end anchored/multiline check string search */
2567 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2569 /* For anchored \G, the only position it can match from is
2570 * (ganch-gofs); we already set startpos to this above; if intuit
2571 * moved us on from there, we can't possibly succeed */
2572 assert(startpos == reginfo->ganch - prog->gofs);
2573 if (s == startpos && regtry(reginfo, &s))
2578 /* Messy cases: unanchored match. */
2579 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2580 /* we have /x+whatever/ */
2581 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2587 if (! prog->anchored_utf8) {
2588 to_utf8_substr(prog);
2590 ch = SvPVX_const(prog->anchored_utf8)[0];
2593 DEBUG_EXECUTE_r( did_match = 1 );
2594 if (regtry(reginfo, &s)) goto got_it;
2596 while (s < strend && *s == ch)
2603 if (! prog->anchored_substr) {
2604 if (! to_byte_substr(prog)) {
2605 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2608 ch = SvPVX_const(prog->anchored_substr)[0];
2611 DEBUG_EXECUTE_r( did_match = 1 );
2612 if (regtry(reginfo, &s)) goto got_it;
2614 while (s < strend && *s == ch)
2619 DEBUG_EXECUTE_r(if (!did_match)
2620 PerlIO_printf(Perl_debug_log,
2621 "Did not find anchored character...\n")
2624 else if (prog->anchored_substr != NULL
2625 || prog->anchored_utf8 != NULL
2626 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2627 && prog->float_max_offset < strend - s)) {
2632 char *last1; /* Last position checked before */
2636 if (prog->anchored_substr || prog->anchored_utf8) {
2638 if (! prog->anchored_utf8) {
2639 to_utf8_substr(prog);
2641 must = prog->anchored_utf8;
2644 if (! prog->anchored_substr) {
2645 if (! to_byte_substr(prog)) {
2646 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2649 must = prog->anchored_substr;
2651 back_max = back_min = prog->anchored_offset;
2654 if (! prog->float_utf8) {
2655 to_utf8_substr(prog);
2657 must = prog->float_utf8;
2660 if (! prog->float_substr) {
2661 if (! to_byte_substr(prog)) {
2662 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2665 must = prog->float_substr;
2667 back_max = prog->float_max_offset;
2668 back_min = prog->float_min_offset;
2674 last = HOP3c(strend, /* Cannot start after this */
2675 -(SSize_t)(CHR_SVLEN(must)
2676 - (SvTAIL(must) != 0) + back_min), strbeg);
2678 if (s > reginfo->strbeg)
2679 last1 = HOPc(s, -1);
2681 last1 = s - 1; /* bogus */
2683 /* XXXX check_substr already used to find "s", can optimize if
2684 check_substr==must. */
2686 strend = HOPc(strend, -dontbother);
2687 while ( (s <= last) &&
2688 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2689 (unsigned char*)strend, must,
2690 multiline ? FBMrf_MULTILINE : 0)) ) {
2691 DEBUG_EXECUTE_r( did_match = 1 );
2692 if (HOPc(s, -back_max) > last1) {
2693 last1 = HOPc(s, -back_min);
2694 s = HOPc(s, -back_max);
2697 char * const t = (last1 >= reginfo->strbeg)
2698 ? HOPc(last1, 1) : last1 + 1;
2700 last1 = HOPc(s, -back_min);
2704 while (s <= last1) {
2705 if (regtry(reginfo, &s))
2708 s++; /* to break out of outer loop */
2715 while (s <= last1) {
2716 if (regtry(reginfo, &s))
2722 DEBUG_EXECUTE_r(if (!did_match) {
2723 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2724 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2725 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2726 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2727 ? "anchored" : "floating"),
2728 quoted, RE_SV_TAIL(must));
2732 else if ( (c = progi->regstclass) ) {
2734 const OPCODE op = OP(progi->regstclass);
2735 /* don't bother with what can't match */
2736 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2737 strend = HOPc(strend, -(minlen - 1));
2740 SV * const prop = sv_newmortal();
2741 regprop(prog, prop, c);
2743 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2745 PerlIO_printf(Perl_debug_log,
2746 "Matching stclass %.*s against %s (%d bytes)\n",
2747 (int)SvCUR(prop), SvPVX_const(prop),
2748 quoted, (int)(strend - s));
2751 if (find_byclass(prog, c, s, strend, reginfo))
2753 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2757 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2765 if (! prog->float_utf8) {
2766 to_utf8_substr(prog);
2768 float_real = prog->float_utf8;
2771 if (! prog->float_substr) {
2772 if (! to_byte_substr(prog)) {
2773 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2776 float_real = prog->float_substr;
2779 little = SvPV_const(float_real, len);
2780 if (SvTAIL(float_real)) {
2781 /* This means that float_real contains an artificial \n on
2782 * the end due to the presence of something like this:
2783 * /foo$/ where we can match both "foo" and "foo\n" at the
2784 * end of the string. So we have to compare the end of the
2785 * string first against the float_real without the \n and
2786 * then against the full float_real with the string. We
2787 * have to watch out for cases where the string might be
2788 * smaller than the float_real or the float_real without
2790 char *checkpos= strend - len;
2792 PerlIO_printf(Perl_debug_log,
2793 "%sChecking for float_real.%s\n",
2794 PL_colors[4], PL_colors[5]));
2795 if (checkpos + 1 < strbeg) {
2796 /* can't match, even if we remove the trailing \n
2797 * string is too short to match */
2799 PerlIO_printf(Perl_debug_log,
2800 "%sString shorter than required trailing substring, cannot match.%s\n",
2801 PL_colors[4], PL_colors[5]));
2803 } else if (memEQ(checkpos + 1, little, len - 1)) {
2804 /* can match, the end of the string matches without the
2806 last = checkpos + 1;
2807 } else if (checkpos < strbeg) {
2808 /* cant match, string is too short when the "\n" is
2811 PerlIO_printf(Perl_debug_log,
2812 "%sString does not contain required trailing substring, cannot match.%s\n",
2813 PL_colors[4], PL_colors[5]));
2815 } else if (!multiline) {
2816 /* non multiline match, so compare with the "\n" at the
2817 * end of the string */
2818 if (memEQ(checkpos, little, len)) {
2822 PerlIO_printf(Perl_debug_log,
2823 "%sString does not contain required trailing substring, cannot match.%s\n",
2824 PL_colors[4], PL_colors[5]));
2828 /* multiline match, so we have to search for a place
2829 * where the full string is located */
2835 last = rninstr(s, strend, little, little + len);
2837 last = strend; /* matching "$" */
2840 /* at one point this block contained a comment which was
2841 * probably incorrect, which said that this was a "should not
2842 * happen" case. Even if it was true when it was written I am
2843 * pretty sure it is not anymore, so I have removed the comment
2844 * and replaced it with this one. Yves */
2846 PerlIO_printf(Perl_debug_log,
2847 "String does not contain required substring, cannot match.\n"
2851 dontbother = strend - last + prog->float_min_offset;
2853 if (minlen && (dontbother < minlen))
2854 dontbother = minlen - 1;
2855 strend -= dontbother; /* this one's always in bytes! */
2856 /* We don't know much -- general case. */
2859 if (regtry(reginfo, &s))
2868 if (regtry(reginfo, &s))
2870 } while (s++ < strend);
2878 /* s/// doesn't like it if $& is earlier than where we asked it to
2879 * start searching (which can happen on something like /.\G/) */
2880 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2881 && (prog->offs[0].start < stringarg - strbeg))
2883 /* this should only be possible under \G */
2884 assert(prog->extflags & RXf_GPOS_SEEN);
2885 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2886 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2892 PerlIO_printf(Perl_debug_log,
2893 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2900 /* clean up; this will trigger destructors that will free all slabs
2901 * above the current one, and cleanup the regmatch_info_aux
2902 * and regmatch_info_aux_eval sructs */
2904 LEAVE_SCOPE(oldsave);
2906 if (RXp_PAREN_NAMES(prog))
2907 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2909 RX_MATCH_UTF8_set(rx, utf8_target);
2911 /* make sure $`, $&, $', and $digit will work later */
2912 if ( !(flags & REXEC_NOT_FIRST) )
2913 S_reg_set_capture_string(aTHX_ rx,
2914 strbeg, reginfo->strend,
2915 sv, flags, utf8_target);
2920 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2921 PL_colors[4], PL_colors[5]));
2923 /* clean up; this will trigger destructors that will free all slabs
2924 * above the current one, and cleanup the regmatch_info_aux
2925 * and regmatch_info_aux_eval sructs */
2927 LEAVE_SCOPE(oldsave);
2930 /* we failed :-( roll it back */
2931 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2932 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2937 Safefree(prog->offs);
2944 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
2945 * Do inc before dec, in case old and new rex are the same */
2946 #define SET_reg_curpm(Re2) \
2947 if (reginfo->info_aux_eval) { \
2948 (void)ReREFCNT_inc(Re2); \
2949 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2950 PM_SETRE((PL_reg_curpm), (Re2)); \
2955 - regtry - try match at specific point
2957 STATIC I32 /* 0 failure, 1 success */
2958 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2962 REGEXP *const rx = reginfo->prog;
2963 regexp *const prog = ReANY(rx);
2965 RXi_GET_DECL(prog,progi);
2966 GET_RE_DEBUG_FLAGS_DECL;
2968 PERL_ARGS_ASSERT_REGTRY;
2970 reginfo->cutpoint=NULL;
2972 prog->offs[0].start = *startposp - reginfo->strbeg;
2973 prog->lastparen = 0;
2974 prog->lastcloseparen = 0;
2976 /* XXXX What this code is doing here?!!! There should be no need
2977 to do this again and again, prog->lastparen should take care of
2980 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2981 * Actually, the code in regcppop() (which Ilya may be meaning by
2982 * prog->lastparen), is not needed at all by the test suite
2983 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2984 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2985 * Meanwhile, this code *is* needed for the
2986 * above-mentioned test suite tests to succeed. The common theme
2987 * on those tests seems to be returning null fields from matches.
2988 * --jhi updated by dapm */
2990 if (prog->nparens) {
2991 regexp_paren_pair *pp = prog->offs;
2993 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3001 result = regmatch(reginfo, *startposp, progi->program + 1);
3003 prog->offs[0].end = result;
3006 if (reginfo->cutpoint)
3007 *startposp= reginfo->cutpoint;
3008 REGCP_UNWIND(lastcp);
3013 #define sayYES goto yes
3014 #define sayNO goto no
3015 #define sayNO_SILENT goto no_silent
3017 /* we dont use STMT_START/END here because it leads to
3018 "unreachable code" warnings, which are bogus, but distracting. */
3019 #define CACHEsayNO \
3020 if (ST.cache_mask) \
3021 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3024 /* this is used to determine how far from the left messages like
3025 'failed...' are printed. It should be set such that messages
3026 are inline with the regop output that created them.
3028 #define REPORT_CODE_OFF 32
3031 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3032 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3033 #define CHRTEST_NOT_A_CP_1 -999
3034 #define CHRTEST_NOT_A_CP_2 -998
3036 /* grab a new slab and return the first slot in it */
3038 STATIC regmatch_state *
3041 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3044 regmatch_slab *s = PL_regmatch_slab->next;
3046 Newx(s, 1, regmatch_slab);
3047 s->prev = PL_regmatch_slab;
3049 PL_regmatch_slab->next = s;
3051 PL_regmatch_slab = s;
3052 return SLAB_FIRST(s);
3056 /* push a new state then goto it */
3058 #define PUSH_STATE_GOTO(state, node, input) \
3059 pushinput = input; \
3061 st->resume_state = state; \
3064 /* push a new state with success backtracking, then goto it */
3066 #define PUSH_YES_STATE_GOTO(state, node, input) \
3067 pushinput = input; \
3069 st->resume_state = state; \
3070 goto push_yes_state;
3077 regmatch() - main matching routine
3079 This is basically one big switch statement in a loop. We execute an op,
3080 set 'next' to point the next op, and continue. If we come to a point which
3081 we may need to backtrack to on failure such as (A|B|C), we push a
3082 backtrack state onto the backtrack stack. On failure, we pop the top
3083 state, and re-enter the loop at the state indicated. If there are no more
3084 states to pop, we return failure.
3086 Sometimes we also need to backtrack on success; for example /A+/, where
3087 after successfully matching one A, we need to go back and try to
3088 match another one; similarly for lookahead assertions: if the assertion
3089 completes successfully, we backtrack to the state just before the assertion
3090 and then carry on. In these cases, the pushed state is marked as
3091 'backtrack on success too'. This marking is in fact done by a chain of
3092 pointers, each pointing to the previous 'yes' state. On success, we pop to
3093 the nearest yes state, discarding any intermediate failure-only states.
3094 Sometimes a yes state is pushed just to force some cleanup code to be
3095 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3096 it to free the inner regex.
3098 Note that failure backtracking rewinds the cursor position, while
3099 success backtracking leaves it alone.
3101 A pattern is complete when the END op is executed, while a subpattern
3102 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3103 ops trigger the "pop to last yes state if any, otherwise return true"
3106 A common convention in this function is to use A and B to refer to the two
3107 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3108 the subpattern to be matched possibly multiple times, while B is the entire
3109 rest of the pattern. Variable and state names reflect this convention.
3111 The states in the main switch are the union of ops and failure/success of
3112 substates associated with with that op. For example, IFMATCH is the op
3113 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3114 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3115 successfully matched A and IFMATCH_A_fail is a state saying that we have
3116 just failed to match A. Resume states always come in pairs. The backtrack
3117 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3118 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3119 on success or failure.
3121 The struct that holds a backtracking state is actually a big union, with
3122 one variant for each major type of op. The variable st points to the
3123 top-most backtrack struct. To make the code clearer, within each
3124 block of code we #define ST to alias the relevant union.
3126 Here's a concrete example of a (vastly oversimplified) IFMATCH
3132 #define ST st->u.ifmatch
3134 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3135 ST.foo = ...; // some state we wish to save
3137 // push a yes backtrack state with a resume value of
3138 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3140 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3143 case IFMATCH_A: // we have successfully executed A; now continue with B
3145 bar = ST.foo; // do something with the preserved value
3148 case IFMATCH_A_fail: // A failed, so the assertion failed
3149 ...; // do some housekeeping, then ...
3150 sayNO; // propagate the failure
3157 For any old-timers reading this who are familiar with the old recursive
3158 approach, the code above is equivalent to:
3160 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3169 ...; // do some housekeeping, then ...
3170 sayNO; // propagate the failure
3173 The topmost backtrack state, pointed to by st, is usually free. If you
3174 want to claim it, populate any ST.foo fields in it with values you wish to
3175 save, then do one of
3177 PUSH_STATE_GOTO(resume_state, node, newinput);
3178 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3180 which sets that backtrack state's resume value to 'resume_state', pushes a
3181 new free entry to the top of the backtrack stack, then goes to 'node'.
3182 On backtracking, the free slot is popped, and the saved state becomes the
3183 new free state. An ST.foo field in this new top state can be temporarily
3184 accessed to retrieve values, but once the main loop is re-entered, it
3185 becomes available for reuse.
3187 Note that the depth of the backtrack stack constantly increases during the
3188 left-to-right execution of the pattern, rather than going up and down with
3189 the pattern nesting. For example the stack is at its maximum at Z at the
3190 end of the pattern, rather than at X in the following:
3192 /(((X)+)+)+....(Y)+....Z/
3194 The only exceptions to this are lookahead/behind assertions and the cut,
3195 (?>A), which pop all the backtrack states associated with A before
3198 Backtrack state structs are allocated in slabs of about 4K in size.
3199 PL_regmatch_state and st always point to the currently active state,
3200 and PL_regmatch_slab points to the slab currently containing
3201 PL_regmatch_state. The first time regmatch() is called, the first slab is
3202 allocated, and is never freed until interpreter destruction. When the slab
3203 is full, a new one is allocated and chained to the end. At exit from
3204 regmatch(), slabs allocated since entry are freed.
3209 #define DEBUG_STATE_pp(pp) \
3211 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3212 PerlIO_printf(Perl_debug_log, \
3213 " %*s"pp" %s%s%s%s%s\n", \
3215 PL_reg_name[st->resume_state], \
3216 ((st==yes_state||st==mark_state) ? "[" : ""), \
3217 ((st==yes_state) ? "Y" : ""), \
3218 ((st==mark_state) ? "M" : ""), \
3219 ((st==yes_state||st==mark_state) ? "]" : "") \
3224 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3229 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3230 const char *start, const char *end, const char *blurb)
3232 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3234 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3239 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3240 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3242 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3243 start, end - start, 60);
3245 PerlIO_printf(Perl_debug_log,
3246 "%s%s REx%s %s against %s\n",
3247 PL_colors[4], blurb, PL_colors[5], s0, s1);
3249 if (utf8_target||utf8_pat)
3250 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3251 utf8_pat ? "pattern" : "",
3252 utf8_pat && utf8_target ? " and " : "",
3253 utf8_target ? "string" : ""
3259 S_dump_exec_pos(pTHX_ const char *locinput,
3260 const regnode *scan,
3261 const char *loc_regeol,
3262 const char *loc_bostr,
3263 const char *loc_reg_starttry,
3264 const bool utf8_target)
3266 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3267 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3268 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3269 /* The part of the string before starttry has one color
3270 (pref0_len chars), between starttry and current
3271 position another one (pref_len - pref0_len chars),
3272 after the current position the third one.
3273 We assume that pref0_len <= pref_len, otherwise we
3274 decrease pref0_len. */
3275 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3276 ? (5 + taill) - l : locinput - loc_bostr;
3279 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3281 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3283 pref0_len = pref_len - (locinput - loc_reg_starttry);
3284 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3285 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3286 ? (5 + taill) - pref_len : loc_regeol - locinput);
3287 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3291 if (pref0_len > pref_len)
3292 pref0_len = pref_len;
3294 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3296 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3297 (locinput - pref_len),pref0_len, 60, 4, 5);
3299 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3300 (locinput - pref_len + pref0_len),
3301 pref_len - pref0_len, 60, 2, 3);
3303 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3304 locinput, loc_regeol - locinput, 10, 0, 1);
3306 const STRLEN tlen=len0+len1+len2;
3307 PerlIO_printf(Perl_debug_log,
3308 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3309 (IV)(locinput - loc_bostr),
3312 (docolor ? "" : "> <"),
3314 (int)(tlen > 19 ? 0 : 19 - tlen),
3321 /* reg_check_named_buff_matched()
3322 * Checks to see if a named buffer has matched. The data array of
3323 * buffer numbers corresponding to the buffer is expected to reside
3324 * in the regexp->data->data array in the slot stored in the ARG() of
3325 * node involved. Note that this routine doesn't actually care about the
3326 * name, that information is not preserved from compilation to execution.
3327 * Returns the index of the leftmost defined buffer with the given name
3328 * or 0 if non of the buffers matched.
3331 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3334 RXi_GET_DECL(rex,rexi);
3335 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3336 I32 *nums=(I32*)SvPVX(sv_dat);
3338 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3340 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3341 if ((I32)rex->lastparen >= nums[n] &&
3342 rex->offs[nums[n]].end != -1)
3352 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3353 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3355 /* This function determines if there are one or two characters that match
3356 * the first character of the passed-in EXACTish node <text_node>, and if
3357 * so, returns them in the passed-in pointers.
3359 * If it determines that no possible character in the target string can
3360 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3361 * the first character in <text_node> requires UTF-8 to represent, and the
3362 * target string isn't in UTF-8.)
3364 * If there are more than two characters that could match the beginning of
3365 * <text_node>, or if more context is required to determine a match or not,
3366 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3368 * The motiviation behind this function is to allow the caller to set up
3369 * tight loops for matching. If <text_node> is of type EXACT, there is
3370 * only one possible character that can match its first character, and so
3371 * the situation is quite simple. But things get much more complicated if
3372 * folding is involved. It may be that the first character of an EXACTFish
3373 * node doesn't participate in any possible fold, e.g., punctuation, so it
3374 * can be matched only by itself. The vast majority of characters that are
3375 * in folds match just two things, their lower and upper-case equivalents.
3376 * But not all are like that; some have multiple possible matches, or match
3377 * sequences of more than one character. This function sorts all that out.
3379 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3380 * loop of trying to match A*, we know we can't exit where the thing
3381 * following it isn't a B. And something can't be a B unless it is the
3382 * beginning of B. By putting a quick test for that beginning in a tight
3383 * loop, we can rule out things that can't possibly be B without having to
3384 * break out of the loop, thus avoiding work. Similarly, if A is a single
3385 * character, we can make a tight loop matching A*, using the outputs of
3388 * If the target string to match isn't in UTF-8, and there aren't
3389 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3390 * the one or two possible octets (which are characters in this situation)
3391 * that can match. In all cases, if there is only one character that can
3392 * match, *<c1p> and *<c2p> will be identical.
3394 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3395 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3396 * can match the beginning of <text_node>. They should be declared with at
3397 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3398 * undefined what these contain.) If one or both of the buffers are
3399 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3400 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3401 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3402 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3403 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3405 const bool utf8_target = reginfo->is_utf8_target;
3407 UV c1 = CHRTEST_NOT_A_CP_1;
3408 UV c2 = CHRTEST_NOT_A_CP_2;
3409 bool use_chrtest_void = FALSE;
3410 const bool is_utf8_pat = reginfo->is_utf8_pat;
3412 /* Used when we have both utf8 input and utf8 output, to avoid converting
3413 * to/from code points */
3414 bool utf8_has_been_setup = FALSE;
3418 U8 *pat = (U8*)STRING(text_node);
3420 if (OP(text_node) == EXACT) {
3422 /* In an exact node, only one thing can be matched, that first
3423 * character. If both the pat and the target are UTF-8, we can just
3424 * copy the input to the output, avoiding finding the code point of
3429 else if (utf8_target) {
3430 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3431 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3432 utf8_has_been_setup = TRUE;
3435 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3438 else /* an EXACTFish node */
3440 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3441 pat + STR_LEN(text_node)))
3443 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3444 pat + STR_LEN(text_node))))
3446 /* Multi-character folds require more context to sort out. Also
3447 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3448 * handled outside this routine */
3449 use_chrtest_void = TRUE;
3451 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3452 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3454 /* Load the folds hash, if not already done */
3456 if (! PL_utf8_foldclosures) {
3457 if (! PL_utf8_tofold) {
3458 U8 dummy[UTF8_MAXBYTES_CASE+1];
3460 /* Force loading this by folding an above-Latin1 char */
3461 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3462 assert(PL_utf8_tofold); /* Verify that worked */
3464 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3467 /* The fold closures data structure is a hash with the keys being
3468 * the UTF-8 of every character that is folded to, like 'k', and
3469 * the values each an array of all code points that fold to its
3470 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3472 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3477 /* Not found in the hash, therefore there are no folds
3478 * containing it, so there is only a single character that
3482 else { /* Does participate in folds */
3483 AV* list = (AV*) *listp;
3484 if (av_len(list) != 1) {
3486 /* If there aren't exactly two folds to this, it is outside
3487 * the scope of this function */
3488 use_chrtest_void = TRUE;
3490 else { /* There are two. Get them */
3491 SV** c_p = av_fetch(list, 0, FALSE);
3493 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3497 c_p = av_fetch(list, 1, FALSE);
3499 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3503 /* Folds that cross the 255/256 boundary are forbidden if
3504 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3505 * pattern character is above 256, and its only other match
3506 * is below 256, the only legal match will be to itself.
3507 * We have thrown away the original, so have to compute
3508 * which is the one above 255 */
3509 if ((c1 < 256) != (c2 < 256)) {
3510 if (OP(text_node) == EXACTFL
3511 || ((OP(text_node) == EXACTFA
3512 || OP(text_node) == EXACTFA_NO_TRIE)
3513 && (isASCII(c1) || isASCII(c2))))
3526 else /* Here, c1 is < 255 */
3528 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3529 && OP(text_node) != EXACTFL
3530 && ((OP(text_node) != EXACTFA
3531 && OP(text_node) != EXACTFA_NO_TRIE)
3534 /* Here, there could be something above Latin1 in the target which
3535 * folds to this character in the pattern. All such cases except
3536 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3537 * involved in their folds, so are outside the scope of this
3539 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3540 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3543 use_chrtest_void = TRUE;
3546 else { /* Here nothing above Latin1 can fold to the pattern character */
3547 switch (OP(text_node)) {
3549 case EXACTFL: /* /l rules */
3550 c2 = PL_fold_locale[c1];
3553 case EXACTF: /* This node only generated for non-utf8
3555 assert(! is_utf8_pat);
3556 if (! utf8_target) { /* /d rules */
3561 /* /u rules for all these. This happens to work for
3562 * EXACTFA as nothing in Latin1 folds to ASCII */
3563 case EXACTFA_NO_TRIE: /* This node only generated for
3564 non-utf8 patterns */
3565 assert(! is_utf8_pat);
3570 c2 = PL_fold_latin1[c1];
3574 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3575 assert(0); /* NOTREACHED */
3580 /* Here have figured things out. Set up the returns */
3581 if (use_chrtest_void) {
3582 *c2p = *c1p = CHRTEST_VOID;
3584 else if (utf8_target) {
3585 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3586 uvchr_to_utf8(c1_utf8, c1);
3587 uvchr_to_utf8(c2_utf8, c2);
3590 /* Invariants are stored in both the utf8 and byte outputs; Use
3591 * negative numbers otherwise for the byte ones. Make sure that the
3592 * byte ones are the same iff the utf8 ones are the same */
3593 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3594 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3597 ? CHRTEST_NOT_A_CP_1
3598 : CHRTEST_NOT_A_CP_2;
3600 else if (c1 > 255) {
3601 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3606 *c1p = *c2p = c2; /* c2 is the only representable value */
3608 else { /* c1 is representable; see about c2 */
3610 *c2p = (c2 < 256) ? c2 : c1;
3616 /* returns -1 on failure, $+[0] on success */
3618 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3620 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3624 const bool utf8_target = reginfo->is_utf8_target;
3625 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3626 REGEXP *rex_sv = reginfo->prog;
3627 regexp *rex = ReANY(rex_sv);
3628 RXi_GET_DECL(rex,rexi);
3629 /* the current state. This is a cached copy of PL_regmatch_state */
3631 /* cache heavy used fields of st in registers */
3634 U32 n = 0; /* general value; init to avoid compiler warning */
3635 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
3636 char *locinput = startpos;
3637 char *pushinput; /* where to continue after a PUSH */
3638 I32 nextchr; /* is always set to UCHARAT(locinput) */
3640 bool result = 0; /* return value of S_regmatch */
3641 int depth = 0; /* depth of backtrack stack */
3642 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3643 const U32 max_nochange_depth =
3644 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3645 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3646 regmatch_state *yes_state = NULL; /* state to pop to on success of
3648 /* mark_state piggy backs on the yes_state logic so that when we unwind
3649 the stack on success we can update the mark_state as we go */
3650 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3651 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3652 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3654 bool no_final = 0; /* prevent failure from backtracking? */
3655 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3656 char *startpoint = locinput;
3657 SV *popmark = NULL; /* are we looking for a mark? */
3658 SV *sv_commit = NULL; /* last mark name seen in failure */
3659 SV *sv_yes_mark = NULL; /* last mark name we have seen
3660 during a successful match */
3661 U32 lastopen = 0; /* last open we saw */
3662 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3663 SV* const oreplsv = GvSVn(PL_replgv);
3664 /* these three flags are set by various ops to signal information to
3665 * the very next op. They have a useful lifetime of exactly one loop
3666 * iteration, and are not preserved or restored by state pushes/pops
3668 bool sw = 0; /* the condition value in (?(cond)a|b) */
3669 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3670 int logical = 0; /* the following EVAL is:
3674 or the following IFMATCH/UNLESSM is:
3675 false: plain (?=foo)
3676 true: used as a condition: (?(?=foo))
3678 PAD* last_pad = NULL;
3680 I32 gimme = G_SCALAR;
3681 CV *caller_cv = NULL; /* who called us */
3682 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3683 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3684 U32 maxopenparen = 0; /* max '(' index seen so far */
3685 int to_complement; /* Invert the result? */
3686 _char_class_number classnum;
3687 bool is_utf8_pat = reginfo->is_utf8_pat;
3690 GET_RE_DEBUG_FLAGS_DECL;
3693 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3694 multicall_oldcatch = 0;
3695 multicall_cv = NULL;
3697 PERL_UNUSED_VAR(multicall_cop);
3698 PERL_UNUSED_VAR(newsp);
3701 PERL_ARGS_ASSERT_REGMATCH;
3703 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3704 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3707 st = PL_regmatch_state;
3709 /* Note that nextchr is a byte even in UTF */
3712 while (scan != NULL) {
3715 SV * const prop = sv_newmortal();
3716 regnode *rnext=regnext(scan);
3717 DUMP_EXEC_POS( locinput, scan, utf8_target );
3718 regprop(rex, prop, scan);
3720 PerlIO_printf(Perl_debug_log,
3721 "%3"IVdf":%*s%s(%"IVdf")\n",
3722 (IV)(scan - rexi->program), depth*2, "",
3724 (PL_regkind[OP(scan)] == END || !rnext) ?
3725 0 : (IV)(rnext - rexi->program));
3728 next = scan + NEXT_OFF(scan);
3731 state_num = OP(scan);
3737 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3739 switch (state_num) {
3740 case BOL: /* /^../ */
3741 if (locinput == reginfo->strbeg)
3745 case MBOL: /* /^../m */
3746 if (locinput == reginfo->strbeg ||
3747 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3753 case SBOL: /* /^../s */
3754 if (locinput == reginfo->strbeg)
3759 if (locinput == reginfo->ganch)
3763 case KEEPS: /* \K */
3764 /* update the startpoint */
3765 st->u.keeper.val = rex->offs[0].start;
3766 rex->offs[0].start = locinput - reginfo->strbeg;
3767 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3768 assert(0); /*NOTREACHED*/
3769 case KEEPS_next_fail:
3770 /* rollback the start point change */
3771 rex->offs[0].start = st->u.keeper.val;
3773 assert(0); /*NOTREACHED*/
3775 case MEOL: /* /..$/m */
3776 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3780 case EOL: /* /..$/ */
3782 case SEOL: /* /..$/s */
3783 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3785 if (reginfo->strend - locinput > 1)
3790 if (!NEXTCHR_IS_EOS)
3794 case SANY: /* /./s */
3797 goto increment_locinput;
3805 case REG_ANY: /* /./ */
3806 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3808 goto increment_locinput;
3812 #define ST st->u.trie
3813 case TRIEC: /* (ab|cd) with known charclass */
3814 /* In this case the charclass data is available inline so
3815 we can fail fast without a lot of extra overhead.
3817 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3819 PerlIO_printf(Perl_debug_log,
3820 "%*s %sfailed to match trie start class...%s\n",
3821 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3824 assert(0); /* NOTREACHED */
3827 case TRIE: /* (ab|cd) */
3828 /* the basic plan of execution of the trie is:
3829 * At the beginning, run though all the states, and
3830 * find the longest-matching word. Also remember the position
3831 * of the shortest matching word. For example, this pattern:
3834 * when matched against the string "abcde", will generate
3835 * accept states for all words except 3, with the longest
3836 * matching word being 4, and the shortest being 2 (with
3837 * the position being after char 1 of the string).
3839 * Then for each matching word, in word order (i.e. 1,2,4,5),
3840 * we run the remainder of the pattern; on each try setting
3841 * the current position to the character following the word,
3842 * returning to try the next word on failure.
3844 * We avoid having to build a list of words at runtime by
3845 * using a compile-time structure, wordinfo[].prev, which
3846 * gives, for each word, the previous accepting word (if any).
3847 * In the case above it would contain the mappings 1->2, 2->0,
3848 * 3->0, 4->5, 5->1. We can use this table to generate, from
3849 * the longest word (4 above), a list of all words, by
3850 * following the list of prev pointers; this gives us the
3851 * unordered list 4,5,1,2. Then given the current word we have
3852 * just tried, we can go through the list and find the
3853 * next-biggest word to try (so if we just failed on word 2,
3854 * the next in the list is 4).
3856 * Since at runtime we don't record the matching position in
3857 * the string for each word, we have to work that out for
3858 * each word we're about to process. The wordinfo table holds
3859 * the character length of each word; given that we recorded
3860 * at the start: the position of the shortest word and its
3861 * length in chars, we just need to move the pointer the
3862 * difference between the two char lengths. Depending on
3863 * Unicode status and folding, that's cheap or expensive.
3865 * This algorithm is optimised for the case where are only a
3866 * small number of accept states, i.e. 0,1, or maybe 2.
3867 * With lots of accepts states, and having to try all of them,
3868 * it becomes quadratic on number of accept states to find all
3873 /* what type of TRIE am I? (utf8 makes this contextual) */
3874 DECL_TRIE_TYPE(scan);
3876 /* what trie are we using right now */
3877 reg_trie_data * const trie
3878 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3879 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3880 U32 state = trie->startstate;
3883 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3885 if (trie->states[ state ].wordnum) {
3887 PerlIO_printf(Perl_debug_log,
3888 "%*s %smatched empty string...%s\n",
3889 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3895 PerlIO_printf(Perl_debug_log,
3896 "%*s %sfailed to match trie start class...%s\n",
3897 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3904 U8 *uc = ( U8* )locinput;
3908 U8 *uscan = (U8*)NULL;
3909 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3910 U32 charcount = 0; /* how many input chars we have matched */
3911 U32 accepted = 0; /* have we seen any accepting states? */
3913 ST.jump = trie->jump;
3916 ST.longfold = FALSE; /* char longer if folded => it's harder */
3919 /* fully traverse the TRIE; note the position of the
3920 shortest accept state and the wordnum of the longest
3923 while ( state && uc <= (U8*)(reginfo->strend) ) {
3924 U32 base = trie->states[ state ].trans.base;
3928 wordnum = trie->states[ state ].wordnum;
3930 if (wordnum) { /* it's an accept state */
3933 /* record first match position */
3935 ST.firstpos = (U8*)locinput;
3940 ST.firstchars = charcount;
3943 if (!ST.nextword || wordnum < ST.nextword)
3944 ST.nextword = wordnum;
3945 ST.topword = wordnum;
3948 DEBUG_TRIE_EXECUTE_r({
3949 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3950 PerlIO_printf( Perl_debug_log,
3951 "%*s %sState: %4"UVxf" Accepted: %c ",
3952 2+depth * 2, "", PL_colors[4],
3953 (UV)state, (accepted ? 'Y' : 'N'));
3956 /* read a char and goto next state */
3957 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
3959 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3960 uscan, len, uvc, charid, foldlen,
3967 base + charid - 1 - trie->uniquecharcount)) >= 0)
3969 && ((U32)offset < trie->lasttrans)
3970 && trie->trans[offset].check == state)
3972 state = trie->trans[offset].next;
3983 DEBUG_TRIE_EXECUTE_r(
3984 PerlIO_printf( Perl_debug_log,
3985 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3986 charid, uvc, (UV)state, PL_colors[5] );
3992 /* calculate total number of accept states */
3997 w = trie->wordinfo[w].prev;
4000 ST.accepted = accepted;
4004 PerlIO_printf( Perl_debug_log,
4005 "%*s %sgot %"IVdf" possible matches%s\n",
4006 REPORT_CODE_OFF + depth * 2, "",
4007 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4009 goto trie_first_try; /* jump into the fail handler */
4011 assert(0); /* NOTREACHED */
4013 case TRIE_next_fail: /* we failed - try next alternative */
4017 REGCP_UNWIND(ST.cp);
4018 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4020 if (!--ST.accepted) {
4022 PerlIO_printf( Perl_debug_log,
4023 "%*s %sTRIE failed...%s\n",
4024 REPORT_CODE_OFF+depth*2, "",
4031 /* Find next-highest word to process. Note that this code
4032 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4035 U16 const nextword = ST.nextword;
4036 reg_trie_wordinfo * const wordinfo
4037 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4038 for (word=ST.topword; word; word=wordinfo[word].prev) {
4039 if (word > nextword && (!min || word < min))
4052 ST.lastparen = rex->lastparen;
4053 ST.lastcloseparen = rex->lastcloseparen;
4057 /* find start char of end of current word */
4059 U32 chars; /* how many chars to skip */
4060 reg_trie_data * const trie
4061 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4063 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4065 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4070 /* the hard option - fold each char in turn and find
4071 * its folded length (which may be different */
4072 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4080 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4088 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4093 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4109 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4110 ? ST.jump[ST.nextword]
4114 PerlIO_printf( Perl_debug_log,
4115 "%*s %sTRIE matched word #%d, continuing%s\n",
4116 REPORT_CODE_OFF+depth*2, "",
4123 if (ST.accepted > 1 || has_cutgroup) {
4124 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4125 assert(0); /* NOTREACHED */
4127 /* only one choice left - just continue */
4129 AV *const trie_words
4130 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4131 SV ** const tmp = av_fetch( trie_words,
4133 SV *sv= tmp ? sv_newmortal() : NULL;
4135 PerlIO_printf( Perl_debug_log,
4136 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4137 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4139 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4140 PL_colors[0], PL_colors[1],
4141 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4143 : "not compiled under -Dr",
4147 locinput = (char*)uc;
4148 continue; /* execute rest of RE */
4149 assert(0); /* NOTREACHED */
4153 case EXACT: { /* /abc/ */
4154 char *s = STRING(scan);
4156 if (utf8_target != is_utf8_pat) {
4157 /* The target and the pattern have differing utf8ness. */
4159 const char * const e = s + ln;
4162 /* The target is utf8, the pattern is not utf8.
4163 * Above-Latin1 code points can't match the pattern;
4164 * invariants match exactly, and the other Latin1 ones need
4165 * to be downgraded to a single byte in order to do the
4166 * comparison. (If we could be confident that the target
4167 * is not malformed, this could be refactored to have fewer
4168 * tests by just assuming that if the first bytes match, it
4169 * is an invariant, but there are tests in the test suite
4170 * dealing with (??{...}) which violate this) */
4172 if (l >= reginfo->strend
4173 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4177 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4184 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4194 /* The target is not utf8, the pattern is utf8. */
4196 if (l >= reginfo->strend
4197 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4201 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4208 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4220 /* The target and the pattern have the same utf8ness. */
4221 /* Inline the first character, for speed. */
4222 if (reginfo->strend - locinput < ln
4223 || UCHARAT(s) != nextchr
4224 || (ln > 1 && memNE(s, locinput, ln)))
4233 case EXACTFL: { /* /abc/il */
4235 const U8 * fold_array;
4237 U32 fold_utf8_flags;
4239 RX_MATCH_TAINTED_on(reginfo->prog);
4240 folder = foldEQ_locale;
4241 fold_array = PL_fold_locale;
4242 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4245 case EXACTFU_SS: /* /\x{df}/iu */
4246 case EXACTFU: /* /abc/iu */
4247 folder = foldEQ_latin1;
4248 fold_array = PL_fold_latin1;
4249 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4252 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4254 assert(! is_utf8_pat);
4256 case EXACTFA: /* /abc/iaa */
4257 folder = foldEQ_latin1;
4258 fold_array = PL_fold_latin1;
4259 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4262 case EXACTF: /* /abc/i This node only generated for
4263 non-utf8 patterns */
4264 assert(! is_utf8_pat);
4266 fold_array = PL_fold;
4267 fold_utf8_flags = 0;
4273 if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) {
4274 /* Either target or the pattern are utf8, or has the issue where
4275 * the fold lengths may differ. */
4276 const char * const l = locinput;
4277 char *e = reginfo->strend;
4279 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4280 l, &e, 0, utf8_target, fold_utf8_flags))
4288 /* Neither the target nor the pattern are utf8 */
4289 if (UCHARAT(s) != nextchr
4291 && UCHARAT(s) != fold_array[nextchr])
4295 if (reginfo->strend - locinput < ln)
4297 if (ln > 1 && ! folder(s, locinput, ln))
4303 /* XXX Could improve efficiency by separating these all out using a
4304 * macro or in-line function. At that point regcomp.c would no longer
4305 * have to set the FLAGS fields of these */
4306 case BOUNDL: /* /\b/l */
4307 case NBOUNDL: /* /\B/l */
4308 RX_MATCH_TAINTED_on(reginfo->prog);
4310 case BOUND: /* /\b/ */
4311 case BOUNDU: /* /\b/u */
4312 case BOUNDA: /* /\b/a */
4313 case NBOUND: /* /\B/ */
4314 case NBOUNDU: /* /\B/u */
4315 case NBOUNDA: /* /\B/a */
4316 /* was last char in word? */
4318 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4319 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4321 if (locinput == reginfo->strbeg)
4324 const U8 * const r =
4325 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4327 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4329 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4330 ln = isWORDCHAR_uni(ln);
4334 LOAD_UTF8_CHARCLASS_ALNUM();
4335 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4340 ln = isWORDCHAR_LC_uvchr(ln);
4341 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4346 /* Here the string isn't utf8, or is utf8 and only ascii
4347 * characters are to match \w. In the latter case looking at
4348 * the byte just prior to the current one may be just the final
4349 * byte of a multi-byte character. This is ok. There are two
4351 * 1) it is a single byte character, and then the test is doing
4352 * just what it's supposed to.
4353 * 2) it is a multi-byte character, in which case the final
4354 * byte is never mistakable for ASCII, and so the test
4355 * will say it is not a word character, which is the
4356 * correct answer. */
4357 ln = (locinput != reginfo->strbeg) ?
4358 UCHARAT(locinput - 1) : '\n';
4359 switch (FLAGS(scan)) {
4360 case REGEX_UNICODE_CHARSET:
4361 ln = isWORDCHAR_L1(ln);
4362 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4364 case REGEX_LOCALE_CHARSET:
4365 ln = isWORDCHAR_LC(ln);
4366 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4368 case REGEX_DEPENDS_CHARSET:
4369 ln = isWORDCHAR(ln);
4370 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4372 case REGEX_ASCII_RESTRICTED_CHARSET:
4373 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4374 ln = isWORDCHAR_A(ln);
4375 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4378 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4382 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4384 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4388 case ANYOF: /* /[abc]/ */
4389 case ANYOF_WARN_SUPER:
4393 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4395 locinput += UTF8SKIP(locinput);
4398 if (!REGINCLASS(rex, scan, (U8*)locinput))
4404 /* The argument (FLAGS) to all the POSIX node types is the class number
4407 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4411 case POSIXL: /* \w or [:punct:] etc. under /l */
4415 /* The locale hasn't influenced the outcome before this, so defer
4416 * tainting until now */
4417 RX_MATCH_TAINTED_on(reginfo->prog);
4419 /* Use isFOO_lc() for characters within Latin1. (Note that
4420 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4421 * wouldn't be invariant) */
4422 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4423 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4427 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4428 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4429 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4430 *(locinput + 1))))))
4435 else { /* Here, must be an above Latin-1 code point */
4436 goto utf8_posix_not_eos;
4439 /* Here, must be utf8 */
4440 locinput += UTF8SKIP(locinput);
4443 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4447 case POSIXD: /* \w or [:punct:] etc. under /d */
4453 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4455 if (NEXTCHR_IS_EOS) {
4459 /* All UTF-8 variants match */
4460 if (! UTF8_IS_INVARIANT(nextchr)) {
4461 goto increment_locinput;
4467 case POSIXA: /* \w or [:punct:] etc. under /a */
4470 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4471 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4472 * character is a single byte */
4475 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4481 /* Here we are either not in utf8, or we matched a utf8-invariant,
4482 * so the next char is the next byte */
4486 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4490 case POSIXU: /* \w or [:punct:] etc. under /u */
4492 if (NEXTCHR_IS_EOS) {
4497 /* Use _generic_isCC() for characters within Latin1. (Note that
4498 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4499 * wouldn't be invariant) */
4500 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4501 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4508 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4509 if (! (to_complement
4510 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4518 else { /* Handle above Latin-1 code points */
4519 classnum = (_char_class_number) FLAGS(scan);
4520 if (classnum < _FIRST_NON_SWASH_CC) {
4522 /* Here, uses a swash to find such code points. Load if if
4523 * not done already */
4524 if (! PL_utf8_swash_ptrs[classnum]) {
4525 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4526 PL_utf8_swash_ptrs[classnum]
4527 = _core_swash_init("utf8",
4528 swash_property_names[classnum],
4529 &PL_sv_undef, 1, 0, NULL, &flags);
4531 if (! (to_complement
4532 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4533 (U8 *) locinput, TRUE))))
4538 else { /* Here, uses macros to find above Latin-1 code points */
4540 case _CC_ENUM_SPACE: /* XXX would require separate
4541 code if we revert the change
4542 of \v matching this */
4543 case _CC_ENUM_PSXSPC:
4544 if (! (to_complement
4545 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4550 case _CC_ENUM_BLANK:
4551 if (! (to_complement
4552 ^ cBOOL(is_HORIZWS_high(locinput))))
4557 case _CC_ENUM_XDIGIT:
4558 if (! (to_complement
4559 ^ cBOOL(is_XDIGIT_high(locinput))))
4564 case _CC_ENUM_VERTSPACE:
4565 if (! (to_complement
4566 ^ cBOOL(is_VERTWS_high(locinput))))
4571 default: /* The rest, e.g. [:cntrl:], can't match
4573 if (! to_complement) {
4579 locinput += UTF8SKIP(locinput);
4583 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4584 a Unicode extended Grapheme Cluster */
4585 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4586 extended Grapheme Cluster is:
4589 | Prepend* Begin Extend*
4592 Begin is: ( Special_Begin | ! Control )
4593 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4594 Extend is: ( Grapheme_Extend | Spacing_Mark )
4595 Control is: [ GCB_Control | CR | LF ]
4596 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4598 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4601 Begin is ( Regular_Begin + Special Begin )
4603 It turns out that 98.4% of all Unicode code points match
4604 Regular_Begin. Doing it this way eliminates a table match in
4605 the previous implementation for almost all Unicode code points.
4607 There is a subtlety with Prepend* which showed up in testing.
4608 Note that the Begin, and only the Begin is required in:
4609 | Prepend* Begin Extend*
4610 Also, Begin contains '! Control'. A Prepend must be a
4611 '! Control', which means it must also be a Begin. What it
4612 comes down to is that if we match Prepend* and then find no
4613 suitable Begin afterwards, that if we backtrack the last
4614 Prepend, that one will be a suitable Begin.
4619 if (! utf8_target) {
4621 /* Match either CR LF or '.', as all the other possibilities
4623 locinput++; /* Match the . or CR */
4624 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4626 && locinput < reginfo->strend
4627 && UCHARAT(locinput) == '\n')
4634 /* Utf8: See if is ( CR LF ); already know that locinput <
4635 * reginfo->strend, so locinput+1 is in bounds */
4636 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4637 && UCHARAT(locinput + 1) == '\n')
4644 /* In case have to backtrack to beginning, then match '.' */
4645 char *starting = locinput;
4647 /* In case have to backtrack the last prepend */
4648 char *previous_prepend = NULL;
4650 LOAD_UTF8_CHARCLASS_GCB();
4652 /* Match (prepend)* */
4653 while (locinput < reginfo->strend
4654 && (len = is_GCB_Prepend_utf8(locinput)))
4656 previous_prepend = locinput;
4660 /* As noted above, if we matched a prepend character, but
4661 * the next thing won't match, back off the last prepend we
4662 * matched, as it is guaranteed to match the begin */
4663 if (previous_prepend
4664 && (locinput >= reginfo->strend
4665 || (! swash_fetch(PL_utf8_X_regular_begin,
4666 (U8*)locinput, utf8_target)
4667 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4670 locinput = previous_prepend;
4673 /* Note that here we know reginfo->strend > locinput, as we
4674 * tested that upon input to this switch case, and if we
4675 * moved locinput forward, we tested the result just above
4676 * and it either passed, or we backed off so that it will
4678 if (swash_fetch(PL_utf8_X_regular_begin,
4679 (U8*)locinput, utf8_target)) {
4680 locinput += UTF8SKIP(locinput);
4682 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4684 /* Here did not match the required 'Begin' in the
4685 * second term. So just match the very first
4686 * character, the '.' of the final term of the regex */
4687 locinput = starting + UTF8SKIP(starting);
4691 /* Here is a special begin. It can be composed of
4692 * several individual characters. One possibility is
4694 if ((len = is_GCB_RI_utf8(locinput))) {
4696 while (locinput < reginfo->strend
4697 && (len = is_GCB_RI_utf8(locinput)))
4701 } else if ((len = is_GCB_T_utf8(locinput))) {
4702 /* Another possibility is T+ */
4704 while (locinput < reginfo->strend
4705 && (len = is_GCB_T_utf8(locinput)))
4711 /* Here, neither RI+ nor T+; must be some other
4712 * Hangul. That means it is one of the others: L,
4713 * LV, LVT or V, and matches:
4714 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4717 while (locinput < reginfo->strend
4718 && (len = is_GCB_L_utf8(locinput)))
4723 /* Here, have exhausted L*. If the next character
4724 * is not an LV, LVT nor V, it means we had to have
4725 * at least one L, so matches L+ in the original
4726 * equation, we have a complete hangul syllable.
4729 if (locinput < reginfo->strend
4730 && is_GCB_LV_LVT_V_utf8(locinput))
4732 /* Otherwise keep going. Must be LV, LVT or V.
4733 * See if LVT, by first ruling out V, then LV */
4734 if (! is_GCB_V_utf8(locinput)
4735 /* All but every TCount one is LV */
4736 && (valid_utf8_to_uvchr((U8 *) locinput,
4741 locinput += UTF8SKIP(locinput);
4744 /* Must be V or LV. Take it, then match
4746 locinput += UTF8SKIP(locinput);
4747 while (locinput < reginfo->strend
4748 && (len = is_GCB_V_utf8(locinput)))
4754 /* And any of LV, LVT, or V can be followed
4756 while (locinput < reginfo->strend
4757 && (len = is_GCB_T_utf8(locinput)))
4765 /* Match any extender */
4766 while (locinput < reginfo->strend
4767 && swash_fetch(PL_utf8_X_extend,
4768 (U8*)locinput, utf8_target))
4770 locinput += UTF8SKIP(locinput);
4774 if (locinput > reginfo->strend) sayNO;
4778 case NREFFL: /* /\g{name}/il */
4779 { /* The capture buffer cases. The ones beginning with N for the
4780 named buffers just convert to the equivalent numbered and
4781 pretend they were called as the corresponding numbered buffer
4783 /* don't initialize these in the declaration, it makes C++
4788 const U8 *fold_array;
4791 RX_MATCH_TAINTED_on(reginfo->prog);
4792 folder = foldEQ_locale;
4793 fold_array = PL_fold_locale;
4795 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4798 case NREFFA: /* /\g{name}/iaa */
4799 folder = foldEQ_latin1;
4800 fold_array = PL_fold_latin1;
4802 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4805 case NREFFU: /* /\g{name}/iu */
4806 folder = foldEQ_latin1;
4807 fold_array = PL_fold_latin1;
4809 utf8_fold_flags = 0;
4812 case NREFF: /* /\g{name}/i */
4814 fold_array = PL_fold;
4816 utf8_fold_flags = 0;
4819 case NREF: /* /\g{name}/ */
4823 utf8_fold_flags = 0;
4826 /* For the named back references, find the corresponding buffer
4828 n = reg_check_named_buff_matched(rex,scan);
4833 goto do_nref_ref_common;
4835 case REFFL: /* /\1/il */
4836 RX_MATCH_TAINTED_on(reginfo->prog);
4837 folder = foldEQ_locale;
4838 fold_array = PL_fold_locale;
4839 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4842 case REFFA: /* /\1/iaa */
4843 folder = foldEQ_latin1;
4844 fold_array = PL_fold_latin1;
4845 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4848 case REFFU: /* /\1/iu */
4849 folder = foldEQ_latin1;
4850 fold_array = PL_fold_latin1;
4851 utf8_fold_flags = 0;
4854 case REFF: /* /\1/i */
4856 fold_array = PL_fold;
4857 utf8_fold_flags = 0;
4860 case REF: /* /\1/ */
4863 utf8_fold_flags = 0;
4867 n = ARG(scan); /* which paren pair */
4870 ln = rex->offs[n].start;
4871 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
4872 if (rex->lastparen < n || ln == -1)
4873 sayNO; /* Do not match unless seen CLOSEn. */
4874 if (ln == rex->offs[n].end)
4877 s = reginfo->strbeg + ln;
4878 if (type != REF /* REF can do byte comparison */
4879 && (utf8_target || type == REFFU))
4880 { /* XXX handle REFFL better */
4881 char * limit = reginfo->strend;
4883 /* This call case insensitively compares the entire buffer
4884 * at s, with the current input starting at locinput, but
4885 * not going off the end given by reginfo->strend, and
4886 * returns in <limit> upon success, how much of the
4887 * current input was matched */
4888 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4889 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4897 /* Not utf8: Inline the first character, for speed. */
4898 if (!NEXTCHR_IS_EOS &&
4899 UCHARAT(s) != nextchr &&
4901 UCHARAT(s) != fold_array[nextchr]))
4903 ln = rex->offs[n].end - ln;
4904 if (locinput + ln > reginfo->strend)
4906 if (ln > 1 && (type == REF
4907 ? memNE(s, locinput, ln)
4908 : ! folder(s, locinput, ln)))
4914 case NOTHING: /* null op; e.g. the 'nothing' following
4915 * the '*' in m{(a+|b)*}' */
4917 case TAIL: /* placeholder while compiling (A|B|C) */
4920 case BACK: /* ??? doesn't appear to be used ??? */
4924 #define ST st->u.eval
4929 regexp_internal *rei;
4930 regnode *startpoint;
4932 case GOSTART: /* (?R) */
4933 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4934 if (cur_eval && cur_eval->locinput==locinput) {
4935 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4936 Perl_croak(aTHX_ "Infinite recursion in regex");
4937 if ( ++nochange_depth > max_nochange_depth )
4939 "Pattern subroutine nesting without pos change"
4940 " exceeded limit in regex");
4947 if (OP(scan)==GOSUB) {
4948 startpoint = scan + ARG2L(scan);
4949 ST.close_paren = ARG(scan);
4951 startpoint = rei->program+1;
4954 goto eval_recurse_doit;
4955 assert(0); /* NOTREACHED */
4957 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4958 if (cur_eval && cur_eval->locinput==locinput) {
4959 if ( ++nochange_depth > max_nochange_depth )
4960 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4965 /* execute the code in the {...} */
4969 OP * const oop = PL_op;
4970 COP * const ocurcop = PL_curcop;
4974 /* save *all* paren positions */
4975 regcppush(rex, 0, maxopenparen);
4976 REGCP_SET(runops_cp);
4979 caller_cv = find_runcv(NULL);
4983 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4985 (REGEXP*)(rexi->data->data[n])
4988 nop = (OP*)rexi->data->data[n+1];
4990 else if (rexi->data->what[n] == 'l') { /* literal code */
4992 nop = (OP*)rexi->data->data[n];
4993 assert(CvDEPTH(newcv));
4996 /* literal with own CV */
4997 assert(rexi->data->what[n] == 'L');
4998 newcv = rex->qr_anoncv;
4999 nop = (OP*)rexi->data->data[n];
5002 /* normally if we're about to execute code from the same
5003 * CV that we used previously, we just use the existing
5004 * CX stack entry. However, its possible that in the
5005 * meantime we may have backtracked, popped from the save
5006 * stack, and undone the SAVECOMPPAD(s) associated with
5007 * PUSH_MULTICALL; in which case PL_comppad no longer
5008 * points to newcv's pad. */
5009 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5011 U8 flags = (CXp_SUB_RE |
5012 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5013 if (last_pushed_cv) {
5014 CHANGE_MULTICALL_FLAGS(newcv, flags);
5017 PUSH_MULTICALL_FLAGS(newcv, flags);
5019 last_pushed_cv = newcv;
5022 /* these assignments are just to silence compiler
5024 multicall_cop = NULL;
5027 last_pad = PL_comppad;
5029 /* the initial nextstate you would normally execute
5030 * at the start of an eval (which would cause error
5031 * messages to come from the eval), may be optimised
5032 * away from the execution path in the regex code blocks;
5033 * so manually set PL_curcop to it initially */
5035 OP *o = cUNOPx(nop)->op_first;
5036 assert(o->op_type == OP_NULL);
5037 if (o->op_targ == OP_SCOPE) {
5038 o = cUNOPo->op_first;
5041 assert(o->op_targ == OP_LEAVE);
5042 o = cUNOPo->op_first;
5043 assert(o->op_type == OP_ENTER);
5047 if (o->op_type != OP_STUB) {
5048 assert( o->op_type == OP_NEXTSTATE
5049 || o->op_type == OP_DBSTATE
5050 || (o->op_type == OP_NULL
5051 && ( o->op_targ == OP_NEXTSTATE
5052 || o->op_targ == OP_DBSTATE
5056 PL_curcop = (COP*)o;
5061 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5062 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5064 rex->offs[0].end = locinput - reginfo->strbeg;
5065 if (reginfo->info_aux_eval->pos_magic)
5066 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5067 reginfo->sv, reginfo->strbeg,
5068 locinput - reginfo->strbeg);
5071 SV *sv_mrk = get_sv("REGMARK", 1);
5072 sv_setsv(sv_mrk, sv_yes_mark);
5075 /* we don't use MULTICALL here as we want to call the
5076 * first op of the block of interest, rather than the
5077 * first op of the sub */
5078 before = (IV)(SP-PL_stack_base);
5080 CALLRUNOPS(aTHX); /* Scalar context. */
5082 if ((IV)(SP-PL_stack_base) == before)
5083 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5089 /* before restoring everything, evaluate the returned
5090 * value, so that 'uninit' warnings don't use the wrong
5091 * PL_op or pad. Also need to process any magic vars
5092 * (e.g. $1) *before* parentheses are restored */
5097 if (logical == 0) /* (?{})/ */
5098 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5099 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5100 sw = cBOOL(SvTRUE(ret));
5103 else { /* /(??{}) */
5104 /* if its overloaded, let the regex compiler handle
5105 * it; otherwise extract regex, or stringify */
5106 if (!SvAMAGIC(ret)) {
5110 if (SvTYPE(sv) == SVt_REGEXP)
5111 re_sv = (REGEXP*) sv;
5112 else if (SvSMAGICAL(sv)) {
5113 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
5115 re_sv = (REGEXP *) mg->mg_obj;
5118 /* force any magic, undef warnings here */
5120 ret = sv_mortalcopy(ret);
5121 (void) SvPV_force_nolen(ret);
5127 /* *** Note that at this point we don't restore
5128 * PL_comppad, (or pop the CxSUB) on the assumption it may
5129 * be used again soon. This is safe as long as nothing
5130 * in the regexp code uses the pad ! */
5132 PL_curcop = ocurcop;
5133 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5134 PL_curpm = PL_reg_curpm;
5140 /* only /(??{})/ from now on */
5143 /* extract RE object from returned value; compiling if
5147 re_sv = reg_temp_copy(NULL, re_sv);
5152 if (SvUTF8(ret) && IN_BYTES) {
5153 /* In use 'bytes': make a copy of the octet
5154 * sequence, but without the flag on */
5156 const char *const p = SvPV(ret, len);
5157 ret = newSVpvn_flags(p, len, SVs_TEMP);
5159 if (rex->intflags & PREGf_USE_RE_EVAL)
5160 pm_flags |= PMf_USE_RE_EVAL;
5162 /* if we got here, it should be an engine which
5163 * supports compiling code blocks and stuff */
5164 assert(rex->engine && rex->engine->op_comp);
5165 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5166 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5167 rex->engine, NULL, NULL,
5168 /* copy /msix etc to inner pattern */
5173 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5175 /* This isn't a first class regexp. Instead, it's
5176 caching a regexp onto an existing, Perl visible
5178 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5180 /* safe to do now that any $1 etc has been
5181 * interpolated into the new pattern string and
5183 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5188 RXp_MATCH_COPIED_off(re);
5189 re->subbeg = rex->subbeg;
5190 re->sublen = rex->sublen;
5191 re->suboffset = rex->suboffset;
5192 re->subcoffset = rex->subcoffset;
5195 debug_start_match(re_sv, utf8_target, locinput,
5196 reginfo->strend, "Matching embedded");
5198 startpoint = rei->program + 1;
5199 ST.close_paren = 0; /* only used for GOSUB */
5201 eval_recurse_doit: /* Share code with GOSUB below this line */
5202 /* run the pattern returned from (??{...}) */
5204 /* Save *all* the positions. */
5205 ST.cp = regcppush(rex, 0, maxopenparen);
5206 REGCP_SET(ST.lastcp);
5209 re->lastcloseparen = 0;
5213 /* invalidate the S-L poscache. We're now executing a
5214 * different set of WHILEM ops (and their associated
5215 * indexes) against the same string, so the bits in the
5216 * cache are meaningless. Setting maxiter to zero forces
5217 * the cache to be invalidated and zeroed before reuse.
5218 * XXX This is too dramatic a measure. Ideally we should
5219 * save the old cache and restore when running the outer
5221 reginfo->poscache_maxiter = 0;
5223 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5225 ST.prev_rex = rex_sv;
5226 ST.prev_curlyx = cur_curlyx;
5228 SET_reg_curpm(rex_sv);
5233 ST.prev_eval = cur_eval;
5235 /* now continue from first node in postoned RE */
5236 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5237 assert(0); /* NOTREACHED */
5240 case EVAL_AB: /* cleanup after a successful (??{A})B */
5241 /* note: this is called twice; first after popping B, then A */
5242 rex_sv = ST.prev_rex;
5243 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5244 SET_reg_curpm(rex_sv);
5245 rex = ReANY(rex_sv);
5246 rexi = RXi_GET(rex);
5248 cur_eval = ST.prev_eval;
5249 cur_curlyx = ST.prev_curlyx;
5251 /* Invalidate cache. See "invalidate" comment above. */
5252 reginfo->poscache_maxiter = 0;
5253 if ( nochange_depth )
5258 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5259 /* note: this is called twice; first after popping B, then A */
5260 rex_sv = ST.prev_rex;
5261 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5262 SET_reg_curpm(rex_sv);
5263 rex = ReANY(rex_sv);
5264 rexi = RXi_GET(rex);
5266 REGCP_UNWIND(ST.lastcp);
5267 regcppop(rex, &maxopenparen);
5268 cur_eval = ST.prev_eval;
5269 cur_curlyx = ST.prev_curlyx;
5270 /* Invalidate cache. See "invalidate" comment above. */
5271 reginfo->poscache_maxiter = 0;
5272 if ( nochange_depth )
5278 n = ARG(scan); /* which paren pair */
5279 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5280 if (n > maxopenparen)
5282 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5283 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5287 (IV)rex->offs[n].start_tmp,
5293 /* XXX really need to log other places start/end are set too */
5294 #define CLOSE_CAPTURE \
5295 rex->offs[n].start = rex->offs[n].start_tmp; \
5296 rex->offs[n].end = locinput - reginfo->strbeg; \
5297 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5298 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5300 PTR2UV(rex->offs), \
5302 (IV)rex->offs[n].start, \
5303 (IV)rex->offs[n].end \
5307 n = ARG(scan); /* which paren pair */
5309 if (n > rex->lastparen)
5311 rex->lastcloseparen = n;
5312 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5317 case ACCEPT: /* (*ACCEPT) */
5321 cursor && OP(cursor)!=END;
5322 cursor=regnext(cursor))
5324 if ( OP(cursor)==CLOSE ){
5326 if ( n <= lastopen ) {
5328 if (n > rex->lastparen)
5330 rex->lastcloseparen = n;
5331 if ( n == ARG(scan) || (cur_eval &&
5332 cur_eval->u.eval.close_paren == n))
5341 case GROUPP: /* (?(1)) */
5342 n = ARG(scan); /* which paren pair */
5343 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5346 case NGROUPP: /* (?(<name>)) */
5347 /* reg_check_named_buff_matched returns 0 for no match */
5348 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5351 case INSUBP: /* (?(R)) */
5353 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5356 case DEFINEP: /* (?(DEFINE)) */
5360 case IFTHEN: /* (?(cond)A|B) */
5361 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5363 next = NEXTOPER(NEXTOPER(scan));
5365 next = scan + ARG(scan);
5366 if (OP(next) == IFTHEN) /* Fake one. */
5367 next = NEXTOPER(NEXTOPER(next));
5371 case LOGICAL: /* modifier for EVAL and IFMATCH */
5372 logical = scan->flags;
5375 /*******************************************************************
5377 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5378 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5379 STAR/PLUS/CURLY/CURLYN are used instead.)
5381 A*B is compiled as <CURLYX><A><WHILEM><B>
5383 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5384 state, which contains the current count, initialised to -1. It also sets
5385 cur_curlyx to point to this state, with any previous value saved in the
5388 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5389 since the pattern may possibly match zero times (i.e. it's a while {} loop
5390 rather than a do {} while loop).
5392 Each entry to WHILEM represents a successful match of A. The count in the
5393 CURLYX block is incremented, another WHILEM state is pushed, and execution
5394 passes to A or B depending on greediness and the current count.
5396 For example, if matching against the string a1a2a3b (where the aN are
5397 substrings that match /A/), then the match progresses as follows: (the
5398 pushed states are interspersed with the bits of strings matched so far):
5401 <CURLYX cnt=0><WHILEM>
5402 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5403 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5404 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5405 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5407 (Contrast this with something like CURLYM, which maintains only a single
5411 a1 <CURLYM cnt=1> a2
5412 a1 a2 <CURLYM cnt=2> a3
5413 a1 a2 a3 <CURLYM cnt=3> b
5416 Each WHILEM state block marks a point to backtrack to upon partial failure
5417 of A or B, and also contains some minor state data related to that
5418 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5419 overall state, such as the count, and pointers to the A and B ops.
5421 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5422 must always point to the *current* CURLYX block, the rules are:
5424 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5425 and set cur_curlyx to point the new block.
5427 When popping the CURLYX block after a successful or unsuccessful match,
5428 restore the previous cur_curlyx.
5430 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5431 to the outer one saved in the CURLYX block.
5433 When popping the WHILEM block after a successful or unsuccessful B match,
5434 restore the previous cur_curlyx.
5436 Here's an example for the pattern (AI* BI)*BO
5437 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5440 curlyx backtrack stack
5441 ------ ---------------
5443 CO <CO prev=NULL> <WO>
5444 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5445 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5446 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5448 At this point the pattern succeeds, and we work back down the stack to
5449 clean up, restoring as we go:
5451 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5452 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5453 CO <CO prev=NULL> <WO>
5456 *******************************************************************/
5458 #define ST st->u.curlyx
5460 case CURLYX: /* start of /A*B/ (for complex A) */
5462 /* No need to save/restore up to this paren */
5463 I32 parenfloor = scan->flags;
5465 assert(next); /* keep Coverity happy */
5466 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5469 /* XXXX Probably it is better to teach regpush to support
5470 parenfloor > maxopenparen ... */
5471 if (parenfloor > (I32)rex->lastparen)
5472 parenfloor = rex->lastparen; /* Pessimization... */
5474 ST.prev_curlyx= cur_curlyx;
5476 ST.cp = PL_savestack_ix;
5478 /* these fields contain the state of the current curly.
5479 * they are accessed by subsequent WHILEMs */
5480 ST.parenfloor = parenfloor;
5485 ST.count = -1; /* this will be updated by WHILEM */
5486 ST.lastloc = NULL; /* this will be updated by WHILEM */
5488 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5489 assert(0); /* NOTREACHED */
5492 case CURLYX_end: /* just finished matching all of A*B */
5493 cur_curlyx = ST.prev_curlyx;
5495 assert(0); /* NOTREACHED */
5497 case CURLYX_end_fail: /* just failed to match all of A*B */
5499 cur_curlyx = ST.prev_curlyx;
5501 assert(0); /* NOTREACHED */
5505 #define ST st->u.whilem
5507 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5509 /* see the discussion above about CURLYX/WHILEM */
5511 int min = ARG1(cur_curlyx->u.curlyx.me);
5512 int max = ARG2(cur_curlyx->u.curlyx.me);
5513 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5515 assert(cur_curlyx); /* keep Coverity happy */
5516 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5517 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5518 ST.cache_offset = 0;
5522 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5523 "%*s whilem: matched %ld out of %d..%d\n",
5524 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5527 /* First just match a string of min A's. */
5530 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5532 cur_curlyx->u.curlyx.lastloc = locinput;
5533 REGCP_SET(ST.lastcp);
5535 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5536 assert(0); /* NOTREACHED */
5539 /* If degenerate A matches "", assume A done. */
5541 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5542 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5543 "%*s whilem: empty match detected, trying continuation...\n",
5544 REPORT_CODE_OFF+depth*2, "")
5546 goto do_whilem_B_max;
5549 /* super-linear cache processing.
5551 * The idea here is that for certain types of CURLYX/WHILEM -
5552 * principally those whose upper bound is infinity (and
5553 * excluding regexes that have things like \1 and other very
5554 * non-regular expresssiony things), then if a pattern like
5555 * /....A*.../ fails and we backtrack to the WHILEM, then we
5556 * make a note that this particular WHILEM op was at string
5557 * position 47 (say) when the rest of pattern failed. Then, if
5558 * we ever find ourselves back at that WHILEM, and at string
5559 * position 47 again, we can just fail immediately rather than
5560 * running the rest of the pattern again.
5562 * This is very handy when patterns start to go
5563 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5564 * with a combinatorial explosion of backtracking.
5566 * The cache is implemented as a bit array, with one bit per
5567 * string byte position per WHILEM op (up to 16) - so its
5568 * between 0.25 and 2x the string size.
5570 * To avoid allocating a poscache buffer every time, we do an
5571 * initially countdown; only after we have executed a WHILEM
5572 * op (string-length x #WHILEMs) times do we allocate the
5575 * The top 4 bits of scan->flags byte say how many different
5576 * relevant CURLLYX/WHILEM op pairs there are, while the
5577 * bottom 4-bits is the identifying index number of this
5583 if (!reginfo->poscache_maxiter) {
5584 /* start the countdown: Postpone detection until we
5585 * know the match is not *that* much linear. */
5586 reginfo->poscache_maxiter
5587 = (reginfo->strend - reginfo->strbeg + 1)
5589 /* possible overflow for long strings and many CURLYX's */
5590 if (reginfo->poscache_maxiter < 0)
5591 reginfo->poscache_maxiter = I32_MAX;
5592 reginfo->poscache_iter = reginfo->poscache_maxiter;
5595 if (reginfo->poscache_iter-- == 0) {
5596 /* initialise cache */
5597 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5598 regmatch_info_aux *const aux = reginfo->info_aux;
5599 if (aux->poscache) {
5600 if ((SSize_t)reginfo->poscache_size < size) {
5601 Renew(aux->poscache, size, char);
5602 reginfo->poscache_size = size;
5604 Zero(aux->poscache, size, char);
5607 reginfo->poscache_size = size;
5608 Newxz(aux->poscache, size, char);
5610 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5611 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5612 PL_colors[4], PL_colors[5])
5616 if (reginfo->poscache_iter < 0) {
5617 /* have we already failed at this position? */
5618 SSize_t offset, mask;
5620 reginfo->poscache_iter = -1; /* stop eventual underflow */
5621 offset = (scan->flags & 0xf) - 1
5622 + (locinput - reginfo->strbeg)
5624 mask = 1 << (offset % 8);
5626 if (reginfo->info_aux->poscache[offset] & mask) {
5627 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5628 "%*s whilem: (cache) already tried at this position...\n",
5629 REPORT_CODE_OFF+depth*2, "")
5631 sayNO; /* cache records failure */
5633 ST.cache_offset = offset;
5634 ST.cache_mask = mask;
5638 /* Prefer B over A for minimal matching. */
5640 if (cur_curlyx->u.curlyx.minmod) {
5641 ST.save_curlyx = cur_curlyx;
5642 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5643 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5645 REGCP_SET(ST.lastcp);
5646 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5648 assert(0); /* NOTREACHED */
5651 /* Prefer A over B for maximal matching. */
5653 if (n < max) { /* More greed allowed? */
5654 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5656 cur_curlyx->u.curlyx.lastloc = locinput;
5657 REGCP_SET(ST.lastcp);
5658 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5659 assert(0); /* NOTREACHED */
5661 goto do_whilem_B_max;
5663 assert(0); /* NOTREACHED */
5665 case WHILEM_B_min: /* just matched B in a minimal match */
5666 case WHILEM_B_max: /* just matched B in a maximal match */
5667 cur_curlyx = ST.save_curlyx;
5669 assert(0); /* NOTREACHED */
5671 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5672 cur_curlyx = ST.save_curlyx;
5673 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5674 cur_curlyx->u.curlyx.count--;
5676 assert(0); /* NOTREACHED */
5678 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5680 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5681 REGCP_UNWIND(ST.lastcp);
5682 regcppop(rex, &maxopenparen);
5683 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5684 cur_curlyx->u.curlyx.count--;
5686 assert(0); /* NOTREACHED */
5688 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5689 REGCP_UNWIND(ST.lastcp);
5690 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5691 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5692 "%*s whilem: failed, trying continuation...\n",
5693 REPORT_CODE_OFF+depth*2, "")
5696 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5697 && ckWARN(WARN_REGEXP)
5698 && !reginfo->warned)
5700 reginfo->warned = TRUE;
5701 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5702 "Complex regular subexpression recursion limit (%d) "
5708 ST.save_curlyx = cur_curlyx;
5709 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5710 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5712 assert(0); /* NOTREACHED */
5714 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5715 cur_curlyx = ST.save_curlyx;
5716 REGCP_UNWIND(ST.lastcp);
5717 regcppop(rex, &maxopenparen);
5719 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5720 /* Maximum greed exceeded */
5721 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5722 && ckWARN(WARN_REGEXP)
5723 && !reginfo->warned)
5725 reginfo->warned = TRUE;
5726 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5727 "Complex regular subexpression recursion "
5728 "limit (%d) exceeded",
5731 cur_curlyx->u.curlyx.count--;
5735 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5736 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5738 /* Try grabbing another A and see if it helps. */
5739 cur_curlyx->u.curlyx.lastloc = locinput;
5740 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5742 REGCP_SET(ST.lastcp);
5743 PUSH_STATE_GOTO(WHILEM_A_min,
5744 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5746 assert(0); /* NOTREACHED */
5749 #define ST st->u.branch
5751 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5752 next = scan + ARG(scan);
5755 scan = NEXTOPER(scan);
5758 case BRANCH: /* /(...|A|...)/ */
5759 scan = NEXTOPER(scan); /* scan now points to inner node */
5760 ST.lastparen = rex->lastparen;
5761 ST.lastcloseparen = rex->lastcloseparen;
5762 ST.next_branch = next;
5765 /* Now go into the branch */
5767 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5769 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5771 assert(0); /* NOTREACHED */
5773 case CUTGROUP: /* /(*THEN)/ */
5774 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5775 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5776 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5777 assert(0); /* NOTREACHED */
5779 case CUTGROUP_next_fail:
5782 if (st->u.mark.mark_name)
5783 sv_commit = st->u.mark.mark_name;
5785 assert(0); /* NOTREACHED */
5789 assert(0); /* NOTREACHED */
5791 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5796 REGCP_UNWIND(ST.cp);
5797 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5798 scan = ST.next_branch;
5799 /* no more branches? */
5800 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5802 PerlIO_printf( Perl_debug_log,
5803 "%*s %sBRANCH failed...%s\n",
5804 REPORT_CODE_OFF+depth*2, "",
5810 continue; /* execute next BRANCH[J] op */
5811 assert(0); /* NOTREACHED */
5813 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5818 #define ST st->u.curlym
5820 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5822 /* This is an optimisation of CURLYX that enables us to push
5823 * only a single backtracking state, no matter how many matches
5824 * there are in {m,n}. It relies on the pattern being constant
5825 * length, with no parens to influence future backrefs
5829 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5831 ST.lastparen = rex->lastparen;
5832 ST.lastcloseparen = rex->lastcloseparen;
5834 /* if paren positive, emulate an OPEN/CLOSE around A */
5836 U32 paren = ST.me->flags;
5837 if (paren > maxopenparen)
5838 maxopenparen = paren;
5839 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5847 ST.c1 = CHRTEST_UNINIT;
5850 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5853 curlym_do_A: /* execute the A in /A{m,n}B/ */
5854 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5855 assert(0); /* NOTREACHED */
5857 case CURLYM_A: /* we've just matched an A */
5859 /* after first match, determine A's length: u.curlym.alen */
5860 if (ST.count == 1) {
5861 if (reginfo->is_utf8_target) {
5862 char *s = st->locinput;
5863 while (s < locinput) {
5869 ST.alen = locinput - st->locinput;
5872 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5875 PerlIO_printf(Perl_debug_log,
5876 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5877 (int)(REPORT_CODE_OFF+(depth*2)), "",
5878 (IV) ST.count, (IV)ST.alen)
5881 if (cur_eval && cur_eval->u.eval.close_paren &&
5882 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5886 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5887 if ( max == REG_INFTY || ST.count < max )
5888 goto curlym_do_A; /* try to match another A */
5890 goto curlym_do_B; /* try to match B */
5892 case CURLYM_A_fail: /* just failed to match an A */
5893 REGCP_UNWIND(ST.cp);
5895 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5896 || (cur_eval && cur_eval->u.eval.close_paren &&
5897 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5900 curlym_do_B: /* execute the B in /A{m,n}B/ */
5901 if (ST.c1 == CHRTEST_UNINIT) {
5902 /* calculate c1 and c2 for possible match of 1st char
5903 * following curly */
5904 ST.c1 = ST.c2 = CHRTEST_VOID;
5905 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5906 regnode *text_node = ST.B;
5907 if (! HAS_TEXT(text_node))
5908 FIND_NEXT_IMPT(text_node);
5911 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5913 But the former is redundant in light of the latter.
5915 if this changes back then the macro for
5916 IS_TEXT and friends need to change.
5918 if (PL_regkind[OP(text_node)] == EXACT) {
5919 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5920 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5930 PerlIO_printf(Perl_debug_log,
5931 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5932 (int)(REPORT_CODE_OFF+(depth*2)),
5935 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5936 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5937 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5938 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5940 /* simulate B failing */
5942 PerlIO_printf(Perl_debug_log,
5943 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
5944 (int)(REPORT_CODE_OFF+(depth*2)),"",
5945 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5946 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5947 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5949 state_num = CURLYM_B_fail;
5950 goto reenter_switch;
5953 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5954 /* simulate B failing */
5956 PerlIO_printf(Perl_debug_log,
5957 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
5958 (int)(REPORT_CODE_OFF+(depth*2)),"",
5959 (int) nextchr, ST.c1, ST.c2)
5961 state_num = CURLYM_B_fail;
5962 goto reenter_switch;
5967 /* emulate CLOSE: mark current A as captured */
5968 I32 paren = ST.me->flags;
5970 rex->offs[paren].start
5971 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
5972 rex->offs[paren].end = locinput - reginfo->strbeg;
5973 if ((U32)paren > rex->lastparen)
5974 rex->lastparen = paren;
5975 rex->lastcloseparen = paren;
5978 rex->offs[paren].end = -1;
5979 if (cur_eval && cur_eval->u.eval.close_paren &&
5980 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5989 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5990 assert(0); /* NOTREACHED */
5992 case CURLYM_B_fail: /* just failed to match a B */
5993 REGCP_UNWIND(ST.cp);
5994 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5996 I32 max = ARG2(ST.me);
5997 if (max != REG_INFTY && ST.count == max)
5999 goto curlym_do_A; /* try to match a further A */
6001 /* backtrack one A */
6002 if (ST.count == ARG1(ST.me) /* min */)
6005 SET_locinput(HOPc(locinput, -ST.alen));
6006 goto curlym_do_B; /* try to match B */
6009 #define ST st->u.curly
6011 #define CURLY_SETPAREN(paren, success) \
6014 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6015 rex->offs[paren].end = locinput - reginfo->strbeg; \
6016 if (paren > rex->lastparen) \
6017 rex->lastparen = paren; \
6018 rex->lastcloseparen = paren; \
6021 rex->offs[paren].end = -1; \
6022 rex->lastparen = ST.lastparen; \
6023 rex->lastcloseparen = ST.lastcloseparen; \
6027 case STAR: /* /A*B/ where A is width 1 char */
6031 scan = NEXTOPER(scan);
6034 case PLUS: /* /A+B/ where A is width 1 char */
6038 scan = NEXTOPER(scan);
6041 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6042 ST.paren = scan->flags; /* Which paren to set */
6043 ST.lastparen = rex->lastparen;
6044 ST.lastcloseparen = rex->lastcloseparen;
6045 if (ST.paren > maxopenparen)
6046 maxopenparen = ST.paren;
6047 ST.min = ARG1(scan); /* min to match */
6048 ST.max = ARG2(scan); /* max to match */
6049 if (cur_eval && cur_eval->u.eval.close_paren &&
6050 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6054 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6057 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6059 ST.min = ARG1(scan); /* min to match */
6060 ST.max = ARG2(scan); /* max to match */
6061 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6064 * Lookahead to avoid useless match attempts
6065 * when we know what character comes next.
6067 * Used to only do .*x and .*?x, but now it allows
6068 * for )'s, ('s and (?{ ... })'s to be in the way
6069 * of the quantifier and the EXACT-like node. -- japhy
6072 assert(ST.min <= ST.max);
6073 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6074 ST.c1 = ST.c2 = CHRTEST_VOID;
6077 regnode *text_node = next;
6079 if (! HAS_TEXT(text_node))
6080 FIND_NEXT_IMPT(text_node);
6082 if (! HAS_TEXT(text_node))
6083 ST.c1 = ST.c2 = CHRTEST_VOID;
6085 if ( PL_regkind[OP(text_node)] != EXACT ) {
6086 ST.c1 = ST.c2 = CHRTEST_VOID;
6090 /* Currently we only get here when
6092 PL_rekind[OP(text_node)] == EXACT
6094 if this changes back then the macro for IS_TEXT and
6095 friends need to change. */
6096 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6097 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6109 char *li = locinput;
6112 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6118 if (ST.c1 == CHRTEST_VOID)
6119 goto curly_try_B_min;
6121 ST.oldloc = locinput;
6123 /* set ST.maxpos to the furthest point along the
6124 * string that could possibly match */
6125 if (ST.max == REG_INFTY) {
6126 ST.maxpos = reginfo->strend - 1;
6128 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6131 else if (utf8_target) {
6132 int m = ST.max - ST.min;
6133 for (ST.maxpos = locinput;
6134 m >0 && ST.maxpos < reginfo->strend; m--)
6135 ST.maxpos += UTF8SKIP(ST.maxpos);
6138 ST.maxpos = locinput + ST.max - ST.min;
6139 if (ST.maxpos >= reginfo->strend)
6140 ST.maxpos = reginfo->strend - 1;
6142 goto curly_try_B_min_known;
6146 /* avoid taking address of locinput, so it can remain
6148 char *li = locinput;
6149 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6150 if (ST.count < ST.min)
6153 if ((ST.count > ST.min)
6154 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6156 /* A{m,n} must come at the end of the string, there's
6157 * no point in backing off ... */
6159 /* ...except that $ and \Z can match before *and* after
6160 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6161 We may back off by one in this case. */
6162 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6166 goto curly_try_B_max;
6168 assert(0); /* NOTREACHED */
6171 case CURLY_B_min_known_fail:
6172 /* failed to find B in a non-greedy match where c1,c2 valid */
6174 REGCP_UNWIND(ST.cp);
6176 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6178 /* Couldn't or didn't -- move forward. */
6179 ST.oldloc = locinput;
6181 locinput += UTF8SKIP(locinput);
6185 curly_try_B_min_known:
6186 /* find the next place where 'B' could work, then call B */
6190 n = (ST.oldloc == locinput) ? 0 : 1;
6191 if (ST.c1 == ST.c2) {
6192 /* set n to utf8_distance(oldloc, locinput) */
6193 while (locinput <= ST.maxpos
6194 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6196 locinput += UTF8SKIP(locinput);
6201 /* set n to utf8_distance(oldloc, locinput) */
6202 while (locinput <= ST.maxpos
6203 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6204 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6206 locinput += UTF8SKIP(locinput);
6211 else { /* Not utf8_target */
6212 if (ST.c1 == ST.c2) {
6213 while (locinput <= ST.maxpos &&
6214 UCHARAT(locinput) != ST.c1)
6218 while (locinput <= ST.maxpos
6219 && UCHARAT(locinput) != ST.c1
6220 && UCHARAT(locinput) != ST.c2)
6223 n = locinput - ST.oldloc;
6225 if (locinput > ST.maxpos)
6228 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6229 * at b; check that everything between oldloc and
6230 * locinput matches */
6231 char *li = ST.oldloc;
6233 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6235 assert(n == REG_INFTY || locinput == li);
6237 CURLY_SETPAREN(ST.paren, ST.count);
6238 if (cur_eval && cur_eval->u.eval.close_paren &&
6239 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6242 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6244 assert(0); /* NOTREACHED */
6247 case CURLY_B_min_fail:
6248 /* failed to find B in a non-greedy match where c1,c2 invalid */
6250 REGCP_UNWIND(ST.cp);
6252 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6254 /* failed -- move forward one */
6256 char *li = locinput;
6257 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6264 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6265 ST.count > 0)) /* count overflow ? */
6268 CURLY_SETPAREN(ST.paren, ST.count);
6269 if (cur_eval && cur_eval->u.eval.close_paren &&
6270 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6273 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6277 assert(0); /* NOTREACHED */
6281 /* a successful greedy match: now try to match B */
6282 if (cur_eval && cur_eval->u.eval.close_paren &&
6283 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6287 bool could_match = locinput < reginfo->strend;
6289 /* If it could work, try it. */
6290 if (ST.c1 != CHRTEST_VOID && could_match) {
6291 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6293 could_match = memEQ(locinput,
6298 UTF8SKIP(locinput));
6301 could_match = UCHARAT(locinput) == ST.c1
6302 || UCHARAT(locinput) == ST.c2;
6305 if (ST.c1 == CHRTEST_VOID || could_match) {
6306 CURLY_SETPAREN(ST.paren, ST.count);
6307 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6308 assert(0); /* NOTREACHED */
6313 case CURLY_B_max_fail:
6314 /* failed to find B in a greedy match */
6316 REGCP_UNWIND(ST.cp);
6318 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6321 if (--ST.count < ST.min)
6323 locinput = HOPc(locinput, -1);
6324 goto curly_try_B_max;
6328 case END: /* last op of main pattern */
6331 /* we've just finished A in /(??{A})B/; now continue with B */
6333 st->u.eval.prev_rex = rex_sv; /* inner */
6335 /* Save *all* the positions. */
6336 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6337 rex_sv = cur_eval->u.eval.prev_rex;
6338 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6339 SET_reg_curpm(rex_sv);
6340 rex = ReANY(rex_sv);
6341 rexi = RXi_GET(rex);
6342 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6344 REGCP_SET(st->u.eval.lastcp);
6346 /* Restore parens of the outer rex without popping the
6348 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6351 st->u.eval.prev_eval = cur_eval;
6352 cur_eval = cur_eval->u.eval.prev_eval;
6354 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6355 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6356 if ( nochange_depth )
6359 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6360 locinput); /* match B */
6363 if (locinput < reginfo->till) {
6364 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6365 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6367 (long)(locinput - startpos),
6368 (long)(reginfo->till - startpos),
6371 sayNO_SILENT; /* Cannot match: too short. */
6373 sayYES; /* Success! */
6375 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6377 PerlIO_printf(Perl_debug_log,
6378 "%*s %ssubpattern success...%s\n",
6379 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6380 sayYES; /* Success! */
6383 #define ST st->u.ifmatch
6388 case SUSPEND: /* (?>A) */
6390 newstart = locinput;
6393 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6395 goto ifmatch_trivial_fail_test;
6397 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6399 ifmatch_trivial_fail_test:
6401 char * const s = HOPBACKc(locinput, scan->flags);
6406 sw = 1 - cBOOL(ST.wanted);
6410 next = scan + ARG(scan);
6418 newstart = locinput;
6422 ST.logical = logical;
6423 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6425 /* execute body of (?...A) */
6426 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6427 assert(0); /* NOTREACHED */
6430 case IFMATCH_A_fail: /* body of (?...A) failed */
6431 ST.wanted = !ST.wanted;
6434 case IFMATCH_A: /* body of (?...A) succeeded */
6436 sw = cBOOL(ST.wanted);
6438 else if (!ST.wanted)
6441 if (OP(ST.me) != SUSPEND) {
6442 /* restore old position except for (?>...) */
6443 locinput = st->locinput;
6445 scan = ST.me + ARG(ST.me);
6448 continue; /* execute B */
6452 case LONGJMP: /* alternative with many branches compiles to
6453 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6454 next = scan + ARG(scan);
6459 case COMMIT: /* (*COMMIT) */
6460 reginfo->cutpoint = reginfo->strend;
6463 case PRUNE: /* (*PRUNE) */
6465 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6466 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6467 assert(0); /* NOTREACHED */
6469 case COMMIT_next_fail:
6473 case OPFAIL: /* (*FAIL) */
6475 assert(0); /* NOTREACHED */
6477 #define ST st->u.mark
6478 case MARKPOINT: /* (*MARK:foo) */
6479 ST.prev_mark = mark_state;
6480 ST.mark_name = sv_commit = sv_yes_mark
6481 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6483 ST.mark_loc = locinput;
6484 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6485 assert(0); /* NOTREACHED */
6487 case MARKPOINT_next:
6488 mark_state = ST.prev_mark;
6490 assert(0); /* NOTREACHED */
6492 case MARKPOINT_next_fail:
6493 if (popmark && sv_eq(ST.mark_name,popmark))
6495 if (ST.mark_loc > startpoint)
6496 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6497 popmark = NULL; /* we found our mark */
6498 sv_commit = ST.mark_name;
6501 PerlIO_printf(Perl_debug_log,
6502 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6503 REPORT_CODE_OFF+depth*2, "",
6504 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6507 mark_state = ST.prev_mark;
6508 sv_yes_mark = mark_state ?
6509 mark_state->u.mark.mark_name : NULL;
6511 assert(0); /* NOTREACHED */
6513 case SKIP: /* (*SKIP) */
6515 /* (*SKIP) : if we fail we cut here*/
6516 ST.mark_name = NULL;
6517 ST.mark_loc = locinput;
6518 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6520 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6521 otherwise do nothing. Meaning we need to scan
6523 regmatch_state *cur = mark_state;
6524 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6527 if ( sv_eq( cur->u.mark.mark_name,
6530 ST.mark_name = find;
6531 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6533 cur = cur->u.mark.prev_mark;
6536 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6539 case SKIP_next_fail:
6541 /* (*CUT:NAME) - Set up to search for the name as we
6542 collapse the stack*/
6543 popmark = ST.mark_name;
6545 /* (*CUT) - No name, we cut here.*/
6546 if (ST.mark_loc > startpoint)
6547 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6548 /* but we set sv_commit to latest mark_name if there
6549 is one so they can test to see how things lead to this
6552 sv_commit=mark_state->u.mark.mark_name;
6556 assert(0); /* NOTREACHED */
6559 case LNBREAK: /* \R */
6560 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6567 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6568 PTR2UV(scan), OP(scan));
6569 Perl_croak(aTHX_ "regexp memory corruption");
6571 /* this is a point to jump to in order to increment
6572 * locinput by one character */
6574 assert(!NEXTCHR_IS_EOS);
6576 locinput += PL_utf8skip[nextchr];
6577 /* locinput is allowed to go 1 char off the end, but not 2+ */
6578 if (locinput > reginfo->strend)
6587 /* switch break jumps here */
6588 scan = next; /* prepare to execute the next op and ... */
6589 continue; /* ... jump back to the top, reusing st */
6590 assert(0); /* NOTREACHED */
6593 /* push a state that backtracks on success */
6594 st->u.yes.prev_yes_state = yes_state;
6598 /* push a new regex state, then continue at scan */
6600 regmatch_state *newst;
6603 regmatch_state *cur = st;
6604 regmatch_state *curyes = yes_state;
6606 regmatch_slab *slab = PL_regmatch_slab;
6607 for (;curd > -1;cur--,curd--) {
6608 if (cur < SLAB_FIRST(slab)) {
6610 cur = SLAB_LAST(slab);
6612 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6613 REPORT_CODE_OFF + 2 + depth * 2,"",
6614 curd, PL_reg_name[cur->resume_state],
6615 (curyes == cur) ? "yes" : ""
6618 curyes = cur->u.yes.prev_yes_state;
6621 DEBUG_STATE_pp("push")
6624 st->locinput = locinput;
6626 if (newst > SLAB_LAST(PL_regmatch_slab))
6627 newst = S_push_slab(aTHX);
6628 PL_regmatch_state = newst;
6630 locinput = pushinput;
6633 assert(0); /* NOTREACHED */
6638 * We get here only if there's trouble -- normally "case END" is
6639 * the terminating point.
6641 Perl_croak(aTHX_ "corrupted regexp pointers");
6647 /* we have successfully completed a subexpression, but we must now
6648 * pop to the state marked by yes_state and continue from there */
6649 assert(st != yes_state);
6651 while (st != yes_state) {
6653 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6654 PL_regmatch_slab = PL_regmatch_slab->prev;
6655 st = SLAB_LAST(PL_regmatch_slab);
6659 DEBUG_STATE_pp("pop (no final)");
6661 DEBUG_STATE_pp("pop (yes)");
6667 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6668 || yes_state > SLAB_LAST(PL_regmatch_slab))
6670 /* not in this slab, pop slab */
6671 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6672 PL_regmatch_slab = PL_regmatch_slab->prev;
6673 st = SLAB_LAST(PL_regmatch_slab);
6675 depth -= (st - yes_state);
6678 yes_state = st->u.yes.prev_yes_state;
6679 PL_regmatch_state = st;
6682 locinput= st->locinput;
6683 state_num = st->resume_state + no_final;
6684 goto reenter_switch;
6687 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6688 PL_colors[4], PL_colors[5]));
6690 if (reginfo->info_aux_eval) {
6691 /* each successfully executed (?{...}) block does the equivalent of
6692 * local $^R = do {...}
6693 * When popping the save stack, all these locals would be undone;
6694 * bypass this by setting the outermost saved $^R to the latest
6696 if (oreplsv != GvSV(PL_replgv))
6697 sv_setsv(oreplsv, GvSV(PL_replgv));
6704 PerlIO_printf(Perl_debug_log,
6705 "%*s %sfailed...%s\n",
6706 REPORT_CODE_OFF+depth*2, "",
6707 PL_colors[4], PL_colors[5])
6719 /* there's a previous state to backtrack to */
6721 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6722 PL_regmatch_slab = PL_regmatch_slab->prev;
6723 st = SLAB_LAST(PL_regmatch_slab);
6725 PL_regmatch_state = st;
6726 locinput= st->locinput;
6728 DEBUG_STATE_pp("pop");
6730 if (yes_state == st)
6731 yes_state = st->u.yes.prev_yes_state;
6733 state_num = st->resume_state + 1; /* failure = success + 1 */
6734 goto reenter_switch;
6739 if (rex->intflags & PREGf_VERBARG_SEEN) {
6740 SV *sv_err = get_sv("REGERROR", 1);
6741 SV *sv_mrk = get_sv("REGMARK", 1);
6743 sv_commit = &PL_sv_no;
6745 sv_yes_mark = &PL_sv_yes;
6748 sv_commit = &PL_sv_yes;
6749 sv_yes_mark = &PL_sv_no;
6751 sv_setsv(sv_err, sv_commit);
6752 sv_setsv(sv_mrk, sv_yes_mark);
6756 if (last_pushed_cv) {
6759 PERL_UNUSED_VAR(SP);
6762 assert(!result || locinput - reginfo->strbeg >= 0);
6763 return result ? locinput - reginfo->strbeg : -1;
6767 - regrepeat - repeatedly match something simple, report how many
6769 * What 'simple' means is a node which can be the operand of a quantifier like
6772 * startposp - pointer a pointer to the start position. This is updated
6773 * to point to the byte following the highest successful
6775 * p - the regnode to be repeatedly matched against.
6776 * reginfo - struct holding match state, such as strend
6777 * max - maximum number of things to match.
6778 * depth - (for debugging) backtracking depth.
6781 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
6782 regmatch_info *const reginfo, I32 max, int depth)
6785 char *scan; /* Pointer to current position in target string */
6787 char *loceol = reginfo->strend; /* local version */
6788 I32 hardcount = 0; /* How many matches so far */
6789 bool utf8_target = reginfo->is_utf8_target;
6790 int to_complement = 0; /* Invert the result? */
6792 _char_class_number classnum;
6794 PERL_UNUSED_ARG(depth);
6797 PERL_ARGS_ASSERT_REGREPEAT;
6800 if (max == REG_INFTY)
6802 else if (! utf8_target && loceol - scan > max)
6803 loceol = scan + max;
6805 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6806 * to the maximum of how far we should go in it (leaving it set to the real
6807 * end, if the maximum permissible would take us beyond that). This allows
6808 * us to make the loop exit condition that we haven't gone past <loceol> to
6809 * also mean that we haven't exceeded the max permissible count, saving a
6810 * test each time through the loop. But it assumes that the OP matches a
6811 * single byte, which is true for most of the OPs below when applied to a
6812 * non-UTF-8 target. Those relatively few OPs that don't have this
6813 * characteristic will have to compensate.
6815 * There is no adjustment for UTF-8 targets, as the number of bytes per
6816 * character varies. OPs will have to test both that the count is less
6817 * than the max permissible (using <hardcount> to keep track), and that we
6818 * are still within the bounds of the string (using <loceol>. A few OPs
6819 * match a single byte no matter what the encoding. They can omit the max
6820 * test if, for the UTF-8 case, they do the adjustment that was skipped
6823 * Thus, the code above sets things up for the common case; and exceptional
6824 * cases need extra work; the common case is to make sure <scan> doesn't
6825 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6826 * count doesn't exceed the maximum permissible */
6831 while (scan < loceol && hardcount < max && *scan != '\n') {
6832 scan += UTF8SKIP(scan);
6836 while (scan < loceol && *scan != '\n')
6842 while (scan < loceol && hardcount < max) {
6843 scan += UTF8SKIP(scan);
6850 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6851 if (utf8_target && loceol - scan > max) {
6853 /* <loceol> hadn't been adjusted in the UTF-8 case */
6861 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6865 /* Can use a simple loop if the pattern char to match on is invariant
6866 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6867 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6868 * true iff it doesn't matter if the argument is in UTF-8 or not */
6869 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
6870 if (utf8_target && loceol - scan > max) {
6871 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6872 * since here, to match at all, 1 char == 1 byte */
6873 loceol = scan + max;
6875 while (scan < loceol && UCHARAT(scan) == c) {
6879 else if (reginfo->is_utf8_pat) {
6881 STRLEN scan_char_len;
6883 /* When both target and pattern are UTF-8, we have to do
6885 while (hardcount < max
6887 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
6888 && memEQ(scan, STRING(p), scan_char_len))
6890 scan += scan_char_len;
6894 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6896 /* Target isn't utf8; convert the character in the UTF-8
6897 * pattern to non-UTF8, and do a simple loop */
6898 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
6899 while (scan < loceol && UCHARAT(scan) == c) {
6902 } /* else pattern char is above Latin1, can't possibly match the
6907 /* Here, the string must be utf8; pattern isn't, and <c> is
6908 * different in utf8 than not, so can't compare them directly.
6909 * Outside the loop, find the two utf8 bytes that represent c, and
6910 * then look for those in sequence in the utf8 string */
6911 U8 high = UTF8_TWO_BYTE_HI(c);
6912 U8 low = UTF8_TWO_BYTE_LO(c);
6914 while (hardcount < max
6915 && scan + 1 < loceol
6916 && UCHARAT(scan) == high
6917 && UCHARAT(scan + 1) == low)
6925 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
6926 assert(! reginfo->is_utf8_pat);
6929 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6933 RXp_MATCH_TAINTED_on(prog);
6934 utf8_flags = FOLDEQ_UTF8_LOCALE;
6937 case EXACTF: /* This node only generated for non-utf8 patterns */
6938 assert(! reginfo->is_utf8_pat);
6944 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6948 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6950 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6952 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
6955 if (c1 == CHRTEST_VOID) {
6956 /* Use full Unicode fold matching */
6957 char *tmpeol = reginfo->strend;
6958 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
6959 while (hardcount < max
6960 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6961 STRING(p), NULL, pat_len,
6962 reginfo->is_utf8_pat, utf8_flags))
6965 tmpeol = reginfo->strend;
6969 else if (utf8_target) {
6971 while (scan < loceol
6973 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6975 scan += UTF8SKIP(scan);
6980 while (scan < loceol
6982 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6983 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6985 scan += UTF8SKIP(scan);
6990 else if (c1 == c2) {
6991 while (scan < loceol && UCHARAT(scan) == c1) {
6996 while (scan < loceol &&
6997 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7006 case ANYOF_WARN_SUPER:
7008 while (hardcount < max
7010 && reginclass(prog, p, (U8*)scan, utf8_target))
7012 scan += UTF8SKIP(scan);
7016 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7021 /* The argument (FLAGS) to all the POSIX node types is the class number */
7028 RXp_MATCH_TAINTED_on(prog);
7029 if (! utf8_target) {
7030 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7036 while (hardcount < max && scan < loceol
7037 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7040 scan += UTF8SKIP(scan);
7053 if (utf8_target && loceol - scan > max) {
7055 /* We didn't adjust <loceol> at the beginning of this routine
7056 * because is UTF-8, but it is actually ok to do so, since here, to
7057 * match, 1 char == 1 byte. */
7058 loceol = scan + max;
7060 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7073 if (! utf8_target) {
7074 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7080 /* The complement of something that matches only ASCII matches all
7081 * UTF-8 variant code points, plus everything in ASCII that isn't
7083 while (hardcount < max && scan < loceol
7084 && (! UTF8_IS_INVARIANT(*scan)
7085 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7087 scan += UTF8SKIP(scan);
7098 if (! utf8_target) {
7099 while (scan < loceol && to_complement
7100 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7107 classnum = (_char_class_number) FLAGS(p);
7108 if (classnum < _FIRST_NON_SWASH_CC) {
7110 /* Here, a swash is needed for above-Latin1 code points.
7111 * Process as many Latin1 code points using the built-in rules.
7112 * Go to another loop to finish processing upon encountering
7113 * the first Latin1 code point. We could do that in this loop
7114 * as well, but the other way saves having to test if the swash
7115 * has been loaded every time through the loop: extra space to
7117 while (hardcount < max && scan < loceol) {
7118 if (UTF8_IS_INVARIANT(*scan)) {
7119 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7126 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7127 if (! (to_complement
7128 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7137 goto found_above_latin1;
7144 /* For these character classes, the knowledge of how to handle
7145 * every code point is compiled in to Perl via a macro. This
7146 * code is written for making the loops as tight as possible.
7147 * It could be refactored to save space instead */
7149 case _CC_ENUM_SPACE: /* XXX would require separate code
7150 if we revert the change of \v
7153 case _CC_ENUM_PSXSPC:
7154 while (hardcount < max
7156 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7158 scan += UTF8SKIP(scan);
7162 case _CC_ENUM_BLANK:
7163 while (hardcount < max
7165 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7167 scan += UTF8SKIP(scan);
7171 case _CC_ENUM_XDIGIT:
7172 while (hardcount < max
7174 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7176 scan += UTF8SKIP(scan);
7180 case _CC_ENUM_VERTSPACE:
7181 while (hardcount < max
7183 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7185 scan += UTF8SKIP(scan);
7189 case _CC_ENUM_CNTRL:
7190 while (hardcount < max
7192 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7194 scan += UTF8SKIP(scan);
7199 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7205 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7207 /* Load the swash if not already present */
7208 if (! PL_utf8_swash_ptrs[classnum]) {
7209 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7210 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7211 "utf8", swash_property_names[classnum],
7212 &PL_sv_undef, 1, 0, NULL, &flags);
7215 while (hardcount < max && scan < loceol
7216 && to_complement ^ cBOOL(_generic_utf8(
7219 swash_fetch(PL_utf8_swash_ptrs[classnum],
7223 scan += UTF8SKIP(scan);
7230 while (hardcount < max && scan < loceol &&
7231 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7236 /* LNBREAK can match one or two latin chars, which is ok, but we
7237 * have to use hardcount in this situation, and throw away the
7238 * adjustment to <loceol> done before the switch statement */
7239 loceol = reginfo->strend;
7240 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7261 /* These are all 0 width, so match right here or not at all. */
7265 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7266 assert(0); /* NOTREACHED */
7273 c = scan - *startposp;
7277 GET_RE_DEBUG_FLAGS_DECL;
7279 SV * const prop = sv_newmortal();
7280 regprop(prog, prop, p);
7281 PerlIO_printf(Perl_debug_log,
7282 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7283 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7291 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7293 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7294 create a copy so that changes the caller makes won't change the shared one.
7295 If <altsvp> is non-null, will return NULL in it, for back-compat.
7298 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7300 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7306 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7311 S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp)
7313 /* Returns the swash for the input 'node' in the regex 'prog'.
7314 * If <doinit> is 'true', will attempt to create the swash if not already
7316 * If <listsvp> is non-null, will return the printable contents of the
7317 * swash. This can be used to get debugging information even before the
7318 * swash exists, by calling this function with 'doinit' set to false, in
7319 * which case the components that will be used to eventually create the
7320 * swash are returned (in a printable form).
7321 * Tied intimately to how regcomp.c sets up the data structure */
7325 SV *si = NULL; /* Input swash initialization string */
7328 RXi_GET_DECL(prog,progi);
7329 const struct reg_data * const data = prog ? progi->data : NULL;
7331 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7333 assert(ANYOF_NONBITMAP(node));
7335 if (data && data->count) {
7336 const U32 n = ARG(node);
7338 if (data->what[n] == 's') {
7339 SV * const rv = MUTABLE_SV(data->data[n]);
7340 AV * const av = MUTABLE_AV(SvRV(rv));
7341 SV **const ary = AvARRAY(av);
7342 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7344 si = *ary; /* ary[0] = the string to initialize the swash with */
7346 /* Elements 2 and 3 are either both present or both absent. [2] is
7347 * any inversion list generated at compile time; [3] indicates if
7348 * that inversion list has any user-defined properties in it. */
7349 if (av_len(av) >= 2) {
7352 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7359 /* Element [1] is reserved for the set-up swash. If already there,
7360 * return it; if not, create it and store it there */
7361 if (ary[1] && SvROK(ary[1])) {
7364 else if (si && doinit) {
7366 sw = _core_swash_init("utf8", /* the utf8 package */
7370 0, /* not from tr/// */
7373 (void)av_store(av, 1, sw);
7378 /* If requested, return a printable version of what this swash matches */
7380 SV* matches_string = newSVpvn("", 0);
7382 /* The swash should be used, if possible, to get the data, as it
7383 * contains the resolved data. But this function can be called at
7384 * compile-time, before everything gets resolved, in which case we
7385 * return the currently best available information, which is the string
7386 * that will eventually be used to do that resolving, 'si' */
7387 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7388 && (si && si != &PL_sv_undef))
7390 sv_catsv(matches_string, si);
7393 /* Add the inversion list to whatever we have. This may have come from
7394 * the swash, or from an input parameter */
7396 sv_catsv(matches_string, _invlist_contents(invlist));
7398 *listsvp = matches_string;
7405 - reginclass - determine if a character falls into a character class
7407 n is the ANYOF regnode
7408 p is the target string
7409 utf8_target tells whether p is in UTF-8.
7411 Returns true if matched; false otherwise.
7413 Note that this can be a synthetic start class, a combination of various
7414 nodes, so things you think might be mutually exclusive, such as locale,
7415 aren't. It can match both locale and non-locale
7420 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target)
7423 const char flags = ANYOF_FLAGS(n);
7427 PERL_ARGS_ASSERT_REGINCLASS;
7429 /* If c is not already the code point, get it. Note that
7430 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7431 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7433 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7434 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7435 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7436 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7437 * UTF8_ALLOW_FFFF */
7438 if (c_len == (STRLEN)-1)
7439 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7442 /* If this character is potentially in the bitmap, check it */
7444 if (ANYOF_BITMAP_TEST(n, c))
7446 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7452 else if (flags & ANYOF_LOCALE) {
7453 RXp_MATCH_TAINTED_on(prog);
7455 if ((flags & ANYOF_LOC_FOLD)
7456 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7460 else if (ANYOF_POSIXL_TEST_ANY_SET(n)) {
7462 /* The data structure is arranged so bits 0, 2, 4, ... are set
7463 * if the class includes the Posix character class given by
7464 * bit/2; and 1, 3, 5, ... are set if the class includes the
7465 * complemented Posix class given by int(bit/2). So we loop
7466 * through the bits, each time changing whether we complement
7467 * the result or not. Suppose for the sake of illustration
7468 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7469 * is set, it means there is a match for this ANYOF node if the
7470 * character is in the class given by the expression (0 / 2 = 0
7471 * = \w). If it is in that class, isFOO_lc() will return 1,
7472 * and since 'to_complement' is 0, the result will stay TRUE,
7473 * and we exit the loop. Suppose instead that bit 0 is 0, but
7474 * bit 1 is 1. That means there is a match if the character
7475 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7476 * but will on bit 1. On the second iteration 'to_complement'
7477 * will be 1, so the exclusive or will reverse things, so we
7478 * are testing for \W. On the third iteration, 'to_complement'
7479 * will be 0, and we would be testing for \s; the fourth
7480 * iteration would test for \S, etc.
7482 * Note that this code assumes that all the classes are closed
7483 * under folding. For example, if a character matches \w, then
7484 * its fold does too; and vice versa. This should be true for
7485 * any well-behaved locale for all the currently defined Posix
7486 * classes, except for :lower: and :upper:, which are handled
7487 * by the pseudo-class :cased: which matches if either of the
7488 * other two does. To get rid of this assumption, an outer
7489 * loop could be used below to iterate over both the source
7490 * character, and its fold (if different) */
7493 int to_complement = 0;
7494 while (count < ANYOF_MAX) {
7495 if (ANYOF_POSIXL_TEST(n, count)
7496 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7508 /* If the bitmap didn't (or couldn't) match, and something outside the
7509 * bitmap could match, try that. Locale nodes specify completely the
7510 * behavior of code points in the bit map (otherwise, a utf8 target would
7511 * cause them to be treated as Unicode and not locale), except in
7512 * the very unlikely event when this node is a synthetic start class, which
7513 * could be a combination of locale and non-locale nodes. So allow locale
7514 * to match for the synthetic start class, which will give a false
7515 * positive that will be resolved when the match is done again as not part
7516 * of the synthetic start class */
7518 if (utf8_target && (flags & ANYOF_ABOVE_LATIN1_ALL) && c >= 256) {
7519 match = TRUE; /* Everything above 255 matches */
7521 else if (ANYOF_NONBITMAP(n)
7522 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7525 || (! (flags & ANYOF_LOCALE))
7526 || OP(n) == ANYOF_SYNTHETIC))))
7528 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7533 } else { /* Convert to utf8 */
7535 utf8_p = bytes_to_utf8(p, &len);
7538 if (swash_fetch(sw, utf8_p, TRUE)) {
7542 /* If we allocated a string above, free it */
7543 if (! utf8_target) Safefree(utf8_p);
7547 if (UNICODE_IS_SUPER(c)
7548 && OP(n) == ANYOF_WARN_SUPER
7549 && ckWARN_d(WARN_NON_UNICODE))
7551 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7552 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7556 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7557 return cBOOL(flags & ANYOF_INVERT) ^ match;
7561 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7563 /* return the position 'off' UTF-8 characters away from 's', forward if
7564 * 'off' >= 0, backwards if negative. But don't go outside of position
7565 * 'lim', which better be < s if off < 0 */
7569 PERL_ARGS_ASSERT_REGHOP3;
7572 while (off-- && s < lim) {
7573 /* XXX could check well-formedness here */
7578 while (off++ && s > lim) {
7580 if (UTF8_IS_CONTINUED(*s)) {
7581 while (s > lim && UTF8_IS_CONTINUATION(*s))
7584 /* XXX could check well-formedness here */
7591 /* there are a bunch of places where we use two reghop3's that should
7592 be replaced with this routine. but since thats not done yet
7593 we ifdef it out - dmq
7596 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7600 PERL_ARGS_ASSERT_REGHOP4;
7603 while (off-- && s < rlim) {
7604 /* XXX could check well-formedness here */
7609 while (off++ && s > llim) {
7611 if (UTF8_IS_CONTINUED(*s)) {
7612 while (s > llim && UTF8_IS_CONTINUATION(*s))
7615 /* XXX could check well-formedness here */
7623 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7627 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7630 while (off-- && s < lim) {
7631 /* XXX could check well-formedness here */
7638 while (off++ && s > lim) {
7640 if (UTF8_IS_CONTINUED(*s)) {
7641 while (s > lim && UTF8_IS_CONTINUATION(*s))
7644 /* XXX could check well-formedness here */
7653 /* when executing a regex that may have (?{}), extra stuff needs setting
7654 up that will be visible to the called code, even before the current
7655 match has finished. In particular:
7657 * $_ is localised to the SV currently being matched;
7658 * pos($_) is created if necessary, ready to be updated on each call-out
7660 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7661 isn't set until the current pattern is successfully finished), so that
7662 $1 etc of the match-so-far can be seen;
7663 * save the old values of subbeg etc of the current regex, and set then
7664 to the current string (again, this is normally only done at the end
7669 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7672 regexp *const rex = ReANY(reginfo->prog);
7673 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7675 eval_state->rex = rex;
7678 /* Make $_ available to executed code. */
7679 if (reginfo->sv != DEFSV) {
7681 DEFSV_set(reginfo->sv);
7684 if (!(mg = mg_find_mglob(reginfo->sv))) {
7685 /* prepare for quick setting of pos */
7686 mg = sv_magicext_mglob(reginfo->sv);
7689 eval_state->pos_magic = mg;
7690 eval_state->pos = mg->mg_len;
7691 eval_state->pos_flags = mg->mg_flags;
7694 eval_state->pos_magic = NULL;
7696 if (!PL_reg_curpm) {
7697 /* PL_reg_curpm is a fake PMOP that we can attach the current
7698 * regex to and point PL_curpm at, so that $1 et al are visible
7699 * within a /(?{})/. It's just allocated once per interpreter the
7700 * first time its needed */
7701 Newxz(PL_reg_curpm, 1, PMOP);
7704 SV* const repointer = &PL_sv_undef;
7705 /* this regexp is also owned by the new PL_reg_curpm, which
7706 will try to free it. */
7707 av_push(PL_regex_padav, repointer);
7708 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
7709 PL_regex_pad = AvARRAY(PL_regex_padav);
7713 SET_reg_curpm(reginfo->prog);
7714 eval_state->curpm = PL_curpm;
7715 PL_curpm = PL_reg_curpm;
7716 if (RXp_MATCH_COPIED(rex)) {
7717 /* Here is a serious problem: we cannot rewrite subbeg,
7718 since it may be needed if this match fails. Thus
7719 $` inside (?{}) could fail... */
7720 eval_state->subbeg = rex->subbeg;
7721 eval_state->sublen = rex->sublen;
7722 eval_state->suboffset = rex->suboffset;
7723 eval_state->subcoffset = rex->subcoffset;
7725 eval_state->saved_copy = rex->saved_copy;
7727 RXp_MATCH_COPIED_off(rex);
7730 eval_state->subbeg = NULL;
7731 rex->subbeg = (char *)reginfo->strbeg;
7733 rex->subcoffset = 0;
7734 rex->sublen = reginfo->strend - reginfo->strbeg;
7738 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
7741 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
7744 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
7745 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
7748 Safefree(aux->poscache);
7752 /* undo the effects of S_setup_eval_state() */
7754 if (eval_state->subbeg) {
7755 regexp * const rex = eval_state->rex;
7756 rex->subbeg = eval_state->subbeg;
7757 rex->sublen = eval_state->sublen;
7758 rex->suboffset = eval_state->suboffset;
7759 rex->subcoffset = eval_state->subcoffset;
7761 rex->saved_copy = eval_state->saved_copy;
7763 RXp_MATCH_COPIED_on(rex);
7765 if (eval_state->pos_magic)
7767 eval_state->pos_magic->mg_len = eval_state->pos;
7768 eval_state->pos_magic->mg_flags =
7769 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
7770 | (eval_state->pos_flags & MGf_BYTES);
7773 PL_curpm = eval_state->curpm;
7776 PL_regmatch_state = aux->old_regmatch_state;
7777 PL_regmatch_slab = aux->old_regmatch_slab;
7779 /* free all slabs above current one - this must be the last action
7780 * of this function, as aux and eval_state are allocated within
7781 * slabs and may be freed here */
7783 s = PL_regmatch_slab->next;
7785 PL_regmatch_slab->next = NULL;
7787 regmatch_slab * const osl = s;
7796 S_to_utf8_substr(pTHX_ regexp *prog)
7798 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7799 * on the converted value */
7803 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7806 if (prog->substrs->data[i].substr
7807 && !prog->substrs->data[i].utf8_substr) {
7808 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7809 prog->substrs->data[i].utf8_substr = sv;
7810 sv_utf8_upgrade(sv);
7811 if (SvVALID(prog->substrs->data[i].substr)) {
7812 if (SvTAIL(prog->substrs->data[i].substr)) {
7813 /* Trim the trailing \n that fbm_compile added last
7815 SvCUR_set(sv, SvCUR(sv) - 1);
7816 /* Whilst this makes the SV technically "invalid" (as its
7817 buffer is no longer followed by "\0") when fbm_compile()
7818 adds the "\n" back, a "\0" is restored. */
7819 fbm_compile(sv, FBMcf_TAIL);
7823 if (prog->substrs->data[i].substr == prog->check_substr)
7824 prog->check_utf8 = sv;
7830 S_to_byte_substr(pTHX_ regexp *prog)
7832 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7833 * on the converted value; returns FALSE if can't be converted. */
7838 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7841 if (prog->substrs->data[i].utf8_substr
7842 && !prog->substrs->data[i].substr) {
7843 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7844 if (! sv_utf8_downgrade(sv, TRUE)) {
7847 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7848 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7849 /* Trim the trailing \n that fbm_compile added last
7851 SvCUR_set(sv, SvCUR(sv) - 1);
7852 fbm_compile(sv, FBMcf_TAIL);
7856 prog->substrs->data[i].substr = sv;
7857 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7858 prog->check_substr = sv;
7867 * c-indentation-style: bsd
7869 * indent-tabs-mode: nil
7872 * ex: set ts=8 sts=4 sw=4 et: